# Diffusers

## Docs

- [번역중](https://huggingface.co/docs/diffusers/main/ko/in_translation.md)
- [설치](https://huggingface.co/docs/diffusers/main/ko/installation.md)
- [훑어보기](https://huggingface.co/docs/diffusers/main/ko/quicktour.md)
- [효과적이고 효율적인 Diffusion](https://huggingface.co/docs/diffusers/main/ko/stable_diffusion.md)
- [Diffusers](https://huggingface.co/docs/diffusers/main/ko/index.md)
- [Stable diffusion XL](https://huggingface.co/docs/diffusers/main/ko/api/pipelines/stable_diffusion/stable_diffusion_xl.md)
- [DreamBooth](https://huggingface.co/docs/diffusers/main/ko/training/dreambooth.md)
- [Low-Rank Adaptation of Large Language Models (LoRA)](https://huggingface.co/docs/diffusers/main/ko/training/lora.md)
- [여러 GPU를 사용한 분산 추론](https://huggingface.co/docs/diffusers/main/ko/training/distributed_inference.md)
- [🧨 Diffusers 학습 예시](https://huggingface.co/docs/diffusers/main/ko/training/overview.md)
- [Text-to-image](https://huggingface.co/docs/diffusers/main/ko/training/text2image.md)
- [Unconditional 이미지 생성](https://huggingface.co/docs/diffusers/main/ko/training/unconditional_training.md)
- [커스텀 Diffusion 학습 예제](https://huggingface.co/docs/diffusers/main/ko/training/custom_diffusion.md)
- [학습을 위한 데이터셋 만들기](https://huggingface.co/docs/diffusers/main/ko/training/create_dataset.md)
- [InstructPix2Pix](https://huggingface.co/docs/diffusers/main/ko/training/instructpix2pix.md)
- [Textual-Inversion](https://huggingface.co/docs/diffusers/main/ko/training/text_inversion.md)
- [ControlNet](https://huggingface.co/docs/diffusers/main/ko/training/controlnet.md)
- [새로운 작업에 대한 모델을 적용하기](https://huggingface.co/docs/diffusers/main/ko/training/adapt_a_model.md)
- [스케줄러](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/schedulers.md)
- [프롬프트에 가중치 부여하기](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/weighted_prompts.md)
- [Text-guided 이미지 인페인팅(inpainting)](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/inpaint.md)
- [Kandinsky](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/kandinsky.md)
- [제어된 생성](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/controlling_generation.md)
- [파이프라인, 모델 및 스케줄러 이해하기](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/write_own_pipeline.md)
- [JAX / Flax에서의 🧨 Stable Diffusion!](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/stable_diffusion_jax_how_to.md)
- [DiffEdit](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/diffedit.md)
- [어댑터 불러오기](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/loading_adapters.md)
- [다양한 Stable Diffusion 포맷 불러오기](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/other-formats.md)
- [파일들을 Hub로 푸시하기](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/push_to_hub.md)
- [조건부 이미지 생성](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/conditional_image_generation.md)
- [텍스트 기반 image-to-image 생성](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/img2img.md)
- [Stable Diffusion XL Turbo](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/sdxl_turbo.md)
- [Stable Video Diffusion](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/svd.md)
- [커스텀 파이프라인 불러오기](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/custom_pipeline_overview.md)
- [Unconditional 이미지 생성](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/unconditional_image_generation.md)
- [파이프라인, 모델, 스케줄러 불러오기](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/loading.md)
- [Shap-E](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/shap-e.md)
- [Text-guided depth-to-image 생성](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/depth2img.md)
- [Textual inversion](https://huggingface.co/docs/diffusers/main/ko/using-diffusers/textual_inversion_inference.md)
- [Diffusion 모델 평가하기[[evaluating-diffusion-models]]](https://huggingface.co/docs/diffusers/main/ko/conceptual/evaluation.md)
- [철학 [[philosophy]]](https://huggingface.co/docs/diffusers/main/ko/conceptual/philosophy.md)
- [Diffusers에 기여하는 방법 🧨 [[how-to-contribute-to-diffusers-]]](https://huggingface.co/docs/diffusers/main/ko/conceptual/contribution.md)
- [🧨 Diffusers의 윤리 지침 [[-diffusers-ethical-guidelines]]](https://huggingface.co/docs/diffusers/main/ko/conceptual/ethical_guidelines.md)
- [Diffusion 모델을 학습하기](https://huggingface.co/docs/diffusers/main/ko/tutorials/basic_training.md)
- [Overview](https://huggingface.co/docs/diffusers/main/ko/tutorials/tutorial_overview.md)
- [Intel Gaudi에서 Stable Diffusion을 사용하는 방법](https://huggingface.co/docs/diffusers/main/ko/optimization/habana.md)
- [Token Merging (토큰 병합)](https://huggingface.co/docs/diffusers/main/ko/optimization/tome.md)
- [xFormers 설치하기](https://huggingface.co/docs/diffusers/main/ko/optimization/xformers.md)
- [메모리와 속도](https://huggingface.co/docs/diffusers/main/ko/optimization/fp16.md)
- [Core ML로 Stable Diffusion을 실행하는 방법](https://huggingface.co/docs/diffusers/main/ko/optimization/coreml.md)
- [Diffusers에서의 PyTorch 2.0 가속화 지원](https://huggingface.co/docs/diffusers/main/ko/optimization/torch2.0.md)
- [Apple Silicon (M1/M2)에서 Stable Diffusion을 사용하는 방법](https://huggingface.co/docs/diffusers/main/ko/optimization/mps.md)
- [추론을 위해 ONNX 런타임을 사용하는 방법](https://huggingface.co/docs/diffusers/main/ko/optimization/onnx.md)
- [추론을 위한 OpenVINO 사용 방법](https://huggingface.co/docs/diffusers/main/ko/optimization/open_vino.md)

### 번역중
https://huggingface.co/docs/diffusers/main/ko/in_translation.md

# 번역중

열심히 번역을 진행중입니다. 조금만 기다려주세요.
감사합니다!

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/in_translation.md" />

### 설치
https://huggingface.co/docs/diffusers/main/ko/installation.md

# 설치

사용하시는 라이브러리에 맞는 🤗 Diffusers를 설치하세요.

🤗 Diffusers는 Python 3.8+, PyTorch 1.7.0+ 및 flax에서 테스트되었습니다. 사용중인 딥러닝 라이브러리에 대한 아래의 설치 안내를 따르세요.

- [PyTorch 설치 안내](https://pytorch.org/get-started/locally/)
- [Flax 설치 안내](https://flax.readthedocs.io/en/latest/)

## pip를 이용한 설치

[가상 환경](https://docs.python.org/3/library/venv.html)에 🤗 Diffusers를 설치해야 합니다.
Python 가상 환경에 익숙하지 않은 경우 [가상환경 pip 설치 가이드](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/)를 살펴보세요.
가상 환경을 사용하면 서로 다른 프로젝트를 더 쉽게 관리하고, 종속성간의 호환성 문제를 피할 수 있습니다.

프로젝트 디렉토리에 가상 환경을 생성하는 것으로 시작하세요:

```bash
python -m venv .env
```

그리고 가상 환경을 활성화합니다:

```bash
source .env/bin/activate
```

이제 다음의 명령어로 🤗 Diffusers를 설치할 준비가 되었습니다:

**PyTorch의 경우**

```bash
pip install diffusers["torch"]
```

**Flax의 경우**

```bash
pip install diffusers["flax"]
```

## 소스로부터 설치

소스에서 `diffusers`를 설치하기 전에, `torch` 및 `accelerate`이 설치되어 있는지 확인하세요.

`torch` 설치에 대해서는 [torch docs](https://pytorch.org/get-started/locally/#start-locally)를 참고하세요.

다음과 같이 `accelerate`을 설치하세요.

```bash
pip install accelerate
```

다음 명령어를 사용하여 소스에서 🤗 Diffusers를 설치하세요:

```bash
pip install git+https://github.com/huggingface/diffusers
```

이 명령어는 최신 `stable` 버전이 아닌 최첨단 `main` 버전을 설치합니다.
`main` 버전은 최신 개발 정보를 최신 상태로 유지하는 데 유용합니다.
예를 들어 마지막 공식 릴리즈 이후 버그가 수정되었지만, 새 릴리즈가 아직 출시되지 않은 경우입니다.
그러나 이는 `main` 버전이 항상 안정적이지 않을 수 있음을 의미합니다.
우리는 `main` 버전이 지속적으로 작동하도록 노력하고 있으며, 대부분의 문제는 보통 몇 시간 또는 하루 안에 해결됩니다.
문제가 발생하면 더 빨리 해결할 수 있도록 [Issue](https://github.com/huggingface/transformers/issues)를 열어주세요!


## 편집가능한 설치

다음을 수행하려면 편집가능한 설치가 필요합니다:

* 소스 코드의 `main` 버전을 사용
* 🤗 Diffusers에 기여 (코드의 변경 사항을 테스트하기 위해 필요)

저장소를 복제하고 다음 명령어를 사용하여 🤗 Diffusers를 설치합니다:

```bash
git clone https://github.com/huggingface/diffusers.git
cd diffusers
```

**PyTorch의 경우**

```sh
pip install -e ".[torch]"
```

**Flax의 경우**

```sh
pip install -e ".[flax]"
```

이러한 명령어들은 저장소를 복제한 폴더와 Python 라이브러리 경로를 연결합니다.
Python은 이제 일반 라이브러리 경로에 더하여 복제한 폴더 내부를 살펴봅니다.
예를들어 Python 패키지가 `~/anaconda3/envs/main/lib/python3.10/site-packages/`에 설치되어 있는 경우 Python은 복제한 폴더인 `~/diffusers/`도 검색합니다.

> [!WARNING]
> 라이브러리를 계속 사용하려면 `diffusers` 폴더를 유지해야 합니다.

이제 다음 명령어를 사용하여 최신 버전의 🤗 Diffusers로 쉽게 업데이트할 수 있습니다:

```bash
cd ~/diffusers/
git pull
```

이렇게 하면, 다음에 실행할 때 Python 환경이 🤗 Diffusers의 `main` 버전을 찾게 됩니다.

## 텔레메트리 로깅에 대한 알림

우리 라이브러리는 `from_pretrained()` 요청 중에 텔레메트리 정보를 원격으로 수집합니다.
이 데이터에는 Diffusers 및 PyTorch/Flax의 버전, 요청된 모델 또는 파이프라인 클래스, 그리고 허브에서 호스팅되는 경우 사전학습된 체크포인트에 대한 경로를 포함합니다.
이 사용 데이터는 문제를 디버깅하고 새로운 기능의 우선순위를 지정하는데 도움이 됩니다.
텔레메트리는 HuggingFace 허브에서 모델과 파이프라인을 불러올 때만 전송되며, 로컬 사용 중에는 수집되지 않습니다.

우리는 추가 정보를 공유하지 않기를 원하는 사람이 있다는 것을 이해하고 개인 정보를 존중하므로, 터미널에서 `DISABLE_TELEMETRY` 환경 변수를 설정하여 텔레메트리 수집을 비활성화할 수 있습니다.

Linux/MacOS에서:
```bash
export DISABLE_TELEMETRY=YES
```

Windows에서:
```bash
set DISABLE_TELEMETRY=YES
```

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/installation.md" />

### 훑어보기
https://huggingface.co/docs/diffusers/main/ko/quicktour.md

# 훑어보기

Diffusion 모델은 이미지나 오디오와 같은 관심 샘플들을 생성하기 위해 랜덤 가우시안 노이즈를 단계별로 제거하도록 학습됩니다. 이로 인해 생성 AI에 대한 관심이 매우 높아졌으며, 인터넷에서 diffusion 생성 이미지의 예를 본 적이 있을 것입니다. 🧨 Diffusers는 누구나 diffusion 모델들을 널리 이용할 수 있도록 하기 위한 라이브러리입니다.

개발자든 일반 사용자든 이 훑어보기를 통해 🧨 Diffusers를 소개하고 빠르게 생성할 수 있도록 도와드립니다! 알아야 할 라이브러리의 주요 구성 요소는 크게 세 가지입니다:

* `DiffusionPipeline`은 추론을 위해 사전 학습된 diffusion 모델에서 샘플을 빠르게 생성하도록 설계된 높은 수준의 엔드투엔드 클래스입니다.
* Diffusion 시스템 생성을 위한 빌딩 블록으로 사용할 수 있는 널리 사용되는 사전 학습된 [model](./api/models) 아키텍처 및 모듈.
* 다양한 [schedulers](./api/schedulers/overview) - 학습을 위해 노이즈를 추가하는 방법과 추론 중에 노이즈 제거된 이미지를 생성하는 방법을 제어하는 알고리즘입니다.

훑어보기에서는 추론을 위해 `DiffusionPipeline`을 사용하는 방법을 보여준 다음, 모델과 스케줄러를 결합하여 `DiffusionPipeline` 내부에서 일어나는 일을 복제하는 방법을 안내합니다.

> [!TIP]
> 훑어보기는 간결한 버전의 🧨 Diffusers 소개로서 [노트북](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/diffusers_intro.ipynb) 빠르게 시작할 수 있도록 도와드립니다. 디퓨저의 목표, 디자인 철학, 핵심 API에 대한 추가 세부 정보를 자세히 알아보려면 노트북을 확인하세요!

시작하기 전에 필요한 라이브러리가 모두 설치되어 있는지 확인하세요:

```py
# 주석 풀어서 Colab에 필요한 라이브러리 설치하기.
#!pip install --upgrade diffusers accelerate transformers
```

- [🤗 Accelerate](https://huggingface.co/docs/accelerate/index)는 추론 및 학습을 위한 모델 로딩 속도를 높여줍니다.
- [🤗 Transformers](https://huggingface.co/docs/transformers/index)는 [Stable Diffusion](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/overview)과 같이 가장 많이 사용되는 diffusion 모델을 실행하는 데 필요합니다.

## DiffusionPipeline

`DiffusionPipeline` 은 추론을 위해 사전 학습된 diffusion 시스템을 사용하는 가장 쉬운 방법입니다. 모델과 스케줄러를 포함하는 엔드 투 엔드 시스템입니다. 다양한 작업에 `DiffusionPipeline`을 바로 사용할 수 있습니다. 아래 표에서 지원되는 몇 가지 작업을 살펴보고, 지원되는 작업의 전체 목록은 [🧨 Diffusers Summary](./api/pipelines/overview#diffusers-summary) 표에서 확인할 수 있습니다.

| **Task**                     | **Description**                                                                                              | **Pipeline**
|------------------------------|--------------------------------------------------------------------------------------------------------------|-----------------|
| Unconditional Image Generation          | generate an image from Gaussian noise | [unconditional_image_generation](./using-diffusers/unconditional_image_generation) |
| Text-Guided Image Generation | generate an image given a text prompt | [conditional_image_generation](./using-diffusers/conditional_image_generation) |
| Text-Guided Image-to-Image Translation     | adapt an image guided by a text prompt | [img2img](./using-diffusers/img2img) |
| Text-Guided Image-Inpainting          | fill the masked part of an image given the image, the mask and a text prompt | [inpaint](./using-diffusers/inpaint) |
| Text-Guided Depth-to-Image Translation | adapt parts of an image guided by a text prompt while preserving structure via depth estimation | [depth2img](./using-diffusers/depth2img) |

먼저 `DiffusionPipeline`의 인스턴스를 생성하고 다운로드할 파이프라인 체크포인트를 지정합니다.
허깅페이스 허브에 저장된 모든 [checkpoint](https://huggingface.co/models?library=diffusers&sort=downloads)에 대해 `DiffusionPipeline`을 사용할 수 있습니다.
이 훑어보기에서는 text-to-image 생성을 위한 [`stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) 체크포인트를 로드합니다.

> [!WARNING]
> [Stable Diffusion](https://huggingface.co/CompVis/stable-diffusion) 모델의 경우, 모델을 실행하기 전에 [라이선스](https://huggingface.co/spaces/CompVis/stable-diffusion-license)를 먼저 주의 깊게 읽어주세요. 🧨 Diffusers는 불쾌하거나 유해한 콘텐츠를 방지하기 위해 [`safety_checker`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py)를 구현하고 있지만, 모델의 향상된 이미지 생성 기능으로 인해 여전히 잠재적으로 유해한 콘텐츠가 생성될 수 있습니다.

`from_pretrained()` 방법으로 모델 로드하기:

```python
>>> from diffusers import DiffusionPipeline

>>> pipeline = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5")
```

The `DiffusionPipeline`은 모든 모델링, 토큰화, 스케줄링 컴포넌트를 다운로드하고 캐시합니다. Stable Diffusion Pipeline은 무엇보다도 `UNet2DConditionModel`과 `PNDMScheduler`로 구성되어 있음을 알 수 있습니다:

```py
>>> pipeline
StableDiffusionPipeline {
  "_class_name": "StableDiffusionPipeline",
  "_diffusers_version": "0.13.1",
  ...,
  "scheduler": [
    "diffusers",
    "PNDMScheduler"
  ],
  ...,
  "unet": [
    "diffusers",
    "UNet2DConditionModel"
  ],
  "vae": [
    "diffusers",
    "AutoencoderKL"
  ]
}
```

이 모델은 약 14억 개의 파라미터로 구성되어 있으므로 GPU에서 파이프라인을 실행할 것을 강력히 권장합니다.
PyTorch에서와 마찬가지로 제너레이터 객체를 GPU로 이동할 수 있습니다:

```python
>>> pipeline.to("cuda")
```

이제 `파이프라인`에 텍스트 프롬프트를 전달하여 이미지를 생성한 다음 노이즈가 제거된 이미지에 액세스할 수 있습니다. 기본적으로 이미지 출력은 [`PIL.Image`](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class) 객체로 감싸집니다.

```python
>>> image = pipeline("An image of a squirrel in Picasso style").images[0]
>>> image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/image_of_squirrel_painting.png"/>
</div>

`save`를 호출하여 이미지를 저장합니다:

```python
>>> image.save("image_of_squirrel_painting.png")
```

### 로컬 파이프라인

파이프라인을 로컬에서 사용할 수도 있습니다. 유일한 차이점은 가중치를 먼저 다운로드해야 한다는 점입니다:

```bash
!git lfs install
!git clone https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5
```

그런 다음 저장된 가중치를 파이프라인에 로드합니다:

```python
>>> pipeline = DiffusionPipeline.from_pretrained("./stable-diffusion-v1-5")
```

이제 위 섹션에서와 같이 파이프라인을 실행할 수 있습니다.

### 스케줄러 교체

스케줄러마다 노이즈 제거 속도와 품질이 서로 다릅니다. 자신에게 가장 적합한 스케줄러를 찾는 가장 좋은 방법은 직접 사용해 보는 것입니다! 🧨 Diffusers의 주요 기능 중 하나는 스케줄러 간에 쉽게 전환이 가능하다는 것입니다. 예를 들어, 기본 스케줄러인 `PNDMScheduler`를 `EulerDiscreteScheduler`로 바꾸려면, `from_config()` 메서드를 사용하여 로드하세요:

```py
>>> from diffusers import EulerDiscreteScheduler

>>> pipeline = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5")
>>> pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)
```

새 스케줄러로 이미지를 생성해보고 어떤 차이가 있는지 확인해 보세요!

다음 섹션에서는 모델과 스케줄러라는 `DiffusionPipeline`을 구성하는 컴포넌트를 자세히 살펴보고 이러한 컴포넌트를 사용하여 고양이 이미지를 생성하는 방법을 배워보겠습니다.

## 모델

대부분의 모델은 노이즈가 있는 샘플을 가져와 각 시간 간격마다 노이즈가 적은 이미지와 입력 이미지 사이의 차이인 *노이즈 잔차*(다른 모델은 이전 샘플을 직접 예측하거나 속도 또는 [`v-prediction`](https://github.com/huggingface/diffusers/blob/5e5ce13e2f89ac45a0066cb3f369462a3cf1d9ef/src/diffusers/schedulers/scheduling_ddim.py#L110)을 예측하는 학습을 합니다)을 예측합니다. 모델을 믹스 앤 매치하여 다른 diffusion 시스템을 만들 수 있습니다.

모델은 `from_pretrained()` 메서드로 시작되며, 이 메서드는 모델 가중치를 로컬에 캐시하여 다음에 모델을 로드할 때 더 빠르게 로드할 수 있습니다. 훑어보기에서는 고양이 이미지에 대해 학습된 체크포인트가 있는 기본적인 unconditional 이미지 생성 모델인 `UNet2DModel`을 로드합니다:

```py
>>> from diffusers import UNet2DModel

>>> repo_id = "google/ddpm-cat-256"
>>> model = UNet2DModel.from_pretrained(repo_id)
```

모델 매개변수에 액세스하려면 `model.config`를 호출합니다:

```py
>>> model.config
```

모델 구성은 🧊 고정된 🧊 딕셔너리로, 모델이 생성된 후에는 해당 매개 변수들을 변경할 수 없습니다. 이는 의도적인 것으로, 처음에 모델 아키텍처를 정의하는 데 사용된 매개변수는 동일하게 유지하면서 다른 매개변수는 추론 중에 조정할 수 있도록 하기 위한 것입니다.

가장 중요한 매개변수들은 다음과 같습니다:

* `sample_size`: 입력 샘플의 높이 및 너비 치수입니다.
* `in_channels`: 입력 샘플의 입력 채널 수입니다.
* `down_block_types` 및 `up_block_types`: UNet 아키텍처를 생성하는 데 사용되는 다운 및 업샘플링 블록의 유형.
* `block_out_channels`: 다운샘플링 블록의 출력 채널 수. 업샘플링 블록의 입력 채널 수에 역순으로 사용되기도 합니다.
* `layers_per_block`: 각 UNet 블록에 존재하는 ResNet 블록의 수입니다.

추론에 모델을 사용하려면 랜덤 가우시안 노이즈로 이미지 모양을 만듭니다. 모델이 여러 개의 무작위 노이즈를 수신할 수 있으므로 'batch' 축, 입력 채널 수에 해당하는 'channel' 축, 이미지의 높이와 너비를 나타내는 'sample_size' 축이 있어야 합니다:

```py
>>> import torch

>>> torch.manual_seed(0)

>>> noisy_sample = torch.randn(1, model.config.in_channels, model.config.sample_size, model.config.sample_size)
>>> noisy_sample.shape
torch.Size([1, 3, 256, 256])
```

추론을 위해 모델에 노이즈가 있는 이미지와 `timestep`을 전달합니다. 'timestep'은 입력 이미지의 노이즈 정도를 나타내며, 시작 부분에 더 많은 노이즈가 있고 끝 부분에 더 적은 노이즈가 있습니다. 이를 통해 모델이 diffusion 과정에서 시작 또는 끝에 더 가까운 위치를 결정할 수 있습니다. `sample` 메서드를 사용하여 모델 출력을 얻습니다:

```py
>>> with torch.no_grad():
...     noisy_residual = model(sample=noisy_sample, timestep=2).sample
```

하지만 실제 예를 생성하려면 노이즈 제거 프로세스를 안내할 스케줄러가 필요합니다. 다음 섹션에서는 모델을 스케줄러와 결합하는 방법에 대해 알아봅니다.

## 스케줄러

스케줄러는 모델 출력이 주어졌을 때 노이즈가 많은 샘플에서 노이즈가 적은 샘플로 전환하는 것을 관리합니다 - 이 경우 'noisy_residual'.

> [!TIP]
> 🧨 Diffusers는 Diffusion 시스템을 구축하기 위한 툴박스입니다. `DiffusionPipeline`을 사용하면 미리 만들어진 Diffusion 시스템을 편리하게 시작할 수 있지만, 모델과 스케줄러 구성 요소를 개별적으로 선택하여 사용자 지정 Diffusion 시스템을 구축할 수도 있습니다.

훑어보기의 경우, `from_config()` 메서드를 사용하여 `DDPMScheduler`를 인스턴스화합니다:

```py
>>> from diffusers import DDPMScheduler

>>> scheduler = DDPMScheduler.from_config(repo_id)
>>> scheduler
DDPMScheduler {
  "_class_name": "DDPMScheduler",
  "_diffusers_version": "0.13.1",
  "beta_end": 0.02,
  "beta_schedule": "linear",
  "beta_start": 0.0001,
  "clip_sample": true,
  "clip_sample_range": 1.0,
  "num_train_timesteps": 1000,
  "prediction_type": "epsilon",
  "trained_betas": null,
  "variance_type": "fixed_small"
}
```

> [!TIP]
> 💡 스케줄러가 구성에서 어떻게 인스턴스화되는지 주목하세요. 모델과 달리 스케줄러에는 학습 가능한 가중치가 없으며 매개변수도 없습니다!

가장 중요한 매개변수는 다음과 같습니다:

* `num_train_timesteps`: 노이즈 제거 프로세스의 길이, 즉 랜덤 가우스 노이즈를 데이터 샘플로 처리하는 데 필요한 타임스텝 수입니다.
* `beta_schedule`: 추론 및 학습에 사용할 노이즈 스케줄 유형입니다.
* `beta_start` 및 `beta_end`: 노이즈 스케줄의 시작 및 종료 노이즈 값입니다.

노이즈가 약간 적은 이미지를 예측하려면 스케줄러의 `step()` 메서드에 모델 출력, `timestep`, 현재 `sample`을 전달하세요.

```py
>>> less_noisy_sample = scheduler.step(model_output=noisy_residual, timestep=2, sample=noisy_sample).prev_sample
>>> less_noisy_sample.shape
```

`less_noisy_sample`을 다음 `timestep`으로 넘기면 노이즈가 더 줄어듭니다! 이제 이 모든 것을 한데 모아 전체 노이즈 제거 과정을 시각화해 보겠습니다.

먼저 노이즈 제거된 이미지를 후처리하여 `PIL.Image`로 표시하는 함수를 만듭니다:

```py
>>> import PIL.Image
>>> import numpy as np


>>> def display_sample(sample, i):
...     image_processed = sample.cpu().permute(0, 2, 3, 1)
...     image_processed = (image_processed + 1.0) * 127.5
...     image_processed = image_processed.numpy().astype(np.uint8)

...     image_pil = PIL.Image.fromarray(image_processed[0])
...     display(f"Image at step {i}")
...     display(image_pil)
```

노이즈 제거 프로세스의 속도를 높이려면 입력과 모델을 GPU로 옮기세요:

```py
>>> model.to("cuda")
>>> noisy_sample = noisy_sample.to("cuda")
```

이제 노이즈가 적은 샘플의 잔차를 예측하고 스케줄러로 노이즈가 적은 샘플을 계산하는 노이즈 제거 루프를 생성합니다:

```py
>>> import tqdm

>>> sample = noisy_sample

>>> for i, t in enumerate(tqdm.tqdm(scheduler.timesteps)):
...     # 1. predict noise residual
...     with torch.no_grad():
...         residual = model(sample, t).sample

...     # 2. compute less noisy image and set x_t -> x_t-1
...     sample = scheduler.step(residual, t, sample).prev_sample

...     # 3. optionally look at image
...     if (i + 1) % 50 == 0:
...         display_sample(sample, i + 1)
```

가만히 앉아서 고양이가 소음으로만 생성되는 것을 지켜보세요!😻

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/diffusion-quicktour.png"/>
</div>

## 다음 단계

이번 훑어보기에서 🧨 Diffusers로 멋진 이미지를 만들어 보셨기를 바랍니다! 다음 단계로 넘어가세요:

* [training](./tutorials/basic_training) 튜토리얼에서 모델을 학습하거나 파인튜닝하여 나만의 이미지를 생성할 수 있습니다.
* 다양한 사용 사례는 공식 및 커뮤니티 [학습 또는 파인튜닝 스크립트](https://github.com/huggingface/diffusers/tree/main/examples#-diffusers-examples) 예시를 참조하세요.
* 스케줄러 로드, 액세스, 변경 및 비교에 대한 자세한 내용은 [다른 스케줄러 사용](./using-diffusers/schedulers) 가이드에서 확인하세요.
* [Stable Diffusion](./stable_diffusion) 가이드에서 프롬프트 엔지니어링, 속도 및 메모리 최적화, 고품질 이미지 생성을 위한 팁과 요령을 살펴보세요.
* [GPU에서 파이토치 최적화](./optimization/fp16) 가이드와 [애플 실리콘(M1/M2)에서의 Stable Diffusion](./optimization/mps) 및 [ONNX 런타임](./optimization/onnx) 실행에 대한 추론 가이드를 통해 🧨 Diffuser 속도를 높이는 방법을 더 자세히 알아보세요.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/quicktour.md" />

### 효과적이고 효율적인 Diffusion
https://huggingface.co/docs/diffusers/main/ko/stable_diffusion.md

# 효과적이고 효율적인 Diffusion


특정 스타일로 이미지를 생성하거나 원하는 내용을 포함하도록`DiffusionPipeline`을 설정하는 것은 까다로울 수 있습니다. 종종 만족스러운 이미지를 얻기까지 `DiffusionPipeline`을 여러 번 실행해야 하는 경우가 많습니다. 그러나 무에서 유를 창조하는 것은 특히 추론을 반복해서 실행하는 경우 계산 집약적인 프로세스입니다.

그렇기 때문에 파이프라인에서 *계산*(속도) 및 *메모리*(GPU RAM) 효율성을 극대화하여 추론 주기 사이의 시간을 단축하여 더 빠르게 반복할 수 있도록 하는 것이 중요합니다.

이 튜토리얼에서는 `DiffusionPipeline`을 사용하여 더 빠르고 효과적으로 생성하는 방법을 안내합니다.

[`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) 모델을 불러와서 시작합니다:

```python
from diffusers import DiffusionPipeline

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipeline = DiffusionPipeline.from_pretrained(model_id)
```

예제 프롬프트는 "portrait of an old warrior chief" 이지만, 자유롭게 자신만의 프롬프트를 사용해도 됩니다:

```python
prompt = "portrait photo of a old warrior chief"
```

## 속도

> [!TIP]
> 💡 GPU에 액세스할 수 없는 경우 다음과 같은 GPU 제공업체에서 무료로 사용할 수 있습니다!. [Colab](https://colab.research.google.com/)

추론 속도를 높이는 가장 간단한 방법 중 하나는 Pytorch 모듈을 사용할 때와 같은 방식으로 GPU에 파이프라인을 배치하는 것입니다:

```python
pipeline = pipeline.to("cuda")
```

동일한 이미지를 사용하고 개선할 수 있는지 확인하려면 [`Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html)를 사용하고 [재현성](./using-diffusers/reusing_seeds)에 대한 시드를 설정하세요:

```python
import torch

generator = torch.Generator("cuda").manual_seed(0)
```

이제 이미지를 생성할 수 있습니다:

```python
image = pipeline(prompt, generator=generator).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_1.png">
</div>

이 프로세스는 T4 GPU에서 약 30초가 소요되었습니다(할당된 GPU가 T4보다 나은 경우 더 빠를 수 있음). 기본적으로 `DiffusionPipeline`은 50개의 추론 단계에 대해 전체 `float32` 정밀도로 추론을 실행합니다. `float16`과 같은 더 낮은 정밀도로 전환하거나 추론 단계를 더 적게 실행하여 속도를 높일 수 있습니다.

`float16`으로 모델을 로드하고 이미지를 생성해 보겠습니다:


```python
import torch

pipeline = DiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipeline = pipeline.to("cuda")
generator = torch.Generator("cuda").manual_seed(0)
image = pipeline(prompt, generator=generator).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_2.png">
</div>

이번에는 이미지를 생성하는 데 약 11초밖에 걸리지 않아 이전보다 3배 가까이 빨라졌습니다!

> [!TIP]
> 💡 파이프라인은 항상 `float16`에서 실행할 것을 강력히 권장하며, 지금까지 출력 품질이 저하되는 경우는 거의 없었습니다.

또 다른 옵션은 추론 단계의 수를 줄이는 것입니다. 보다 효율적인 스케줄러를 선택하면 출력 품질 저하 없이 단계 수를 줄이는 데 도움이 될 수 있습니다. 현재 모델과 호환되는 스케줄러는 `compatibles` 메서드를 호출하여 `DiffusionPipeline`에서 찾을 수 있습니다:

```python
pipeline.scheduler.compatibles
[
    diffusers.schedulers.scheduling_lms_discrete.LMSDiscreteScheduler,
    diffusers.schedulers.scheduling_unipc_multistep.UniPCMultistepScheduler,
    diffusers.schedulers.scheduling_k_dpm_2_discrete.KDPM2DiscreteScheduler,
    diffusers.schedulers.scheduling_deis_multistep.DEISMultistepScheduler,
    diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler,
    diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler,
    diffusers.schedulers.scheduling_ddpm.DDPMScheduler,
    diffusers.schedulers.scheduling_dpmsolver_singlestep.DPMSolverSinglestepScheduler,
    diffusers.schedulers.scheduling_k_dpm_2_ancestral_discrete.KDPM2AncestralDiscreteScheduler,
    diffusers.schedulers.scheduling_heun_discrete.HeunDiscreteScheduler,
    diffusers.schedulers.scheduling_pndm.PNDMScheduler,
    diffusers.schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteScheduler,
    diffusers.schedulers.scheduling_ddim.DDIMScheduler,
]
```

Stable Diffusion 모델은 일반적으로 약 50개의 추론 단계가 필요한 `PNDMScheduler`를 기본으로 사용하지만, `DPMSolverMultistepScheduler`와 같이 성능이 더 뛰어난 스케줄러는 약 20개 또는 25개의 추론 단계만 필요로 합니다. 새 스케줄러를 로드하려면 `ConfigMixin.from_config()` 메서드를 사용합니다:

```python
from diffusers import DPMSolverMultistepScheduler

pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)
```

`num_inference_steps`를 20으로 설정합니다:

```python
generator = torch.Generator("cuda").manual_seed(0)
image = pipeline(prompt, generator=generator, num_inference_steps=20).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_3.png">
</div>

추론시간을 4초로 단축할 수 있었습니다! ⚡️

## 메모리

파이프라인 성능 향상의 또 다른 핵심은 메모리 사용량을 줄이는 것인데, 초당 생성되는 이미지 수를 최대화하려고 하는 경우가 많기 때문에 간접적으로 더 빠른 속도를 의미합니다. 한 번에 생성할 수 있는 이미지 수를 확인하는 가장 쉬운 방법은 `OutOfMemoryError`(OOM)이 발생할 때까지 다양한 배치 크기를 시도해 보는 것입니다.

프롬프트 목록과 `Generators`에서 이미지 배치를 생성하는 함수를 만듭니다. 좋은 결과를 생성하는 경우 재사용할 수 있도록 각 `Generator`에 시드를 할당해야 합니다.

```python
def get_inputs(batch_size=1):
    generator = [torch.Generator("cuda").manual_seed(i) for i in range(batch_size)]
    prompts = batch_size * [prompt]
    num_inference_steps = 20

    return {"prompt": prompts, "generator": generator, "num_inference_steps": num_inference_steps}
```

또한 각 이미지 배치를 보여주는 기능이 필요합니다:

```python
from PIL import Image


def image_grid(imgs, rows=2, cols=2):
    w, h = imgs[0].size
    grid = Image.new("RGB", size=(cols * w, rows * h))

    for i, img in enumerate(imgs):
        grid.paste(img, box=(i % cols * w, i // cols * h))
    return grid
```

`batch_size=4`부터 시작해 얼마나 많은 메모리를 소비했는지 확인합니다:

```python
images = pipeline(**get_inputs(batch_size=4)).images
image_grid(images)
```

RAM이 더 많은 GPU가 아니라면 위의 코드에서 `OOM` 오류가 반환되었을 것입니다! 대부분의 메모리는 cross-attention 레이어가 차지합니다. 이 작업을 배치로 실행하는 대신 순차적으로 실행하면 상당한 양의 메모리를 절약할 수 있습니다. 파이프라인을 구성하여 `enable_attention_slicing()` 함수를 사용하기만 하면 됩니다:


```python
pipeline.enable_attention_slicing()
```

이제 `batch_size`를 8로 늘려보세요!

```python
images = pipeline(**get_inputs(batch_size=8)).images
image_grid(images, rows=2, cols=4)
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_5.png">
</div>

이전에는 4개의 이미지를 배치로 생성할 수도 없었지만, 이제는 이미지당 약 3.5초 만에 8개의 이미지를 배치로 생성할 수 있습니다! 이는 아마도 품질 저하 없이 T4 GPU에서 가장 빠른 속도일 것입니다.

## 품질

지난 두 섹션에서는 `fp16`을 사용하여 파이프라인의 속도를 최적화하고, 더 성능이 좋은 스케줄러를 사용하여 추론 단계의 수를 줄이고, attention slicing을 활성화하여 메모리 소비를 줄이는 방법을 배웠습니다. 이제 생성된 이미지의 품질을 개선하는 방법에 대해 집중적으로 알아보겠습니다.


### 더 나은 체크포인트

가장 확실한 단계는 더 나은 체크포인트를 사용하는 것입니다. Stable Diffusion 모델은 좋은 출발점이며, 공식 출시 이후 몇 가지 개선된 버전도 출시되었습니다. 하지만 최신 버전을 사용한다고 해서 자동으로 더 나은 결과를 얻을 수 있는 것은 아닙니다. 여전히 다양한 체크포인트를 직접 실험해보고, [negative prompts](https://minimaxir.com/2022/11/stable-diffusion-negative-prompt/) 사용 등 약간의 조사를 통해 최상의 결과를 얻어야 합니다.

이 분야가 성장함에 따라 특정 스타일을 연출할 수 있도록 세밀하게 조정된 고품질 체크포인트가 점점 더 많아지고 있습니다. [Hub](https://huggingface.co/models?library=diffusers&sort=downloads)와 [Diffusers Gallery](https://huggingface.co/spaces/huggingface-projects/diffusers-gallery)를 둘러보고 관심 있는 것을 찾아보세요!


### 더 나은 파이프라인 구성 요소

현재 파이프라인 구성 요소를 최신 버전으로 교체해 볼 수도 있습니다. Stability AI의 최신 [autodecoder](https://huggingface.co/stabilityai/stable-diffusion-2-1/tree/main/vae)를 파이프라인에 로드하고 몇 가지 이미지를 생성해 보겠습니다:


```python
from diffusers import AutoencoderKL

vae = AutoencoderKL.from_pretrained("stabilityai/sd-vae-ft-mse", torch_dtype=torch.float16).to("cuda")
pipeline.vae = vae
images = pipeline(**get_inputs(batch_size=8)).images
image_grid(images, rows=2, cols=4)
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_6.png">
</div>

### 더 나은 프롬프트 엔지니어링

이미지를 생성하는 데 사용하는 텍스트 프롬프트는 *prompt engineering*이라고 할 정도로 매우 중요합니다. 프롬프트 엔지니어링 시 고려해야 할 몇 가지 사항은 다음과 같습니다:

- 생성하려는 이미지 또는 유사한 이미지가 인터넷에 어떻게 저장되어 있는가?
- 내가 원하는 스타일로 모델을 유도하기 위해 어떤 추가 세부 정보를 제공할 수 있는가?

이를 염두에 두고 색상과 더 높은 품질의 디테일을 포함하도록 프롬프트를 개선해 봅시다:


```python
prompt += ", tribal panther make up, blue on red, side profile, looking away, serious eyes"
prompt += " 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta"
```

새로운 프롬프트로 이미지 배치를 생성합니다:

```python
images = pipeline(**get_inputs(batch_size=8)).images
image_grid(images, rows=2, cols=4)
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_7.png">
</div>

꽤 인상적입니다! `1`의 시드를 가진 `Generator`에 해당하는 두 번째 이미지에 피사체의 나이에 대한 텍스트를 추가하여 조금 더 조정해 보겠습니다:

```python
prompts = [
    "portrait photo of the oldest warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
    "portrait photo of a old warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
    "portrait photo of a warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
    "portrait photo of a young warrior chief, tribal panther make up, blue on red, side profile, looking away, serious eyes 50mm portrait photography, hard rim lighting photography--beta --ar 2:3  --beta --upbeta",
]

generator = [torch.Generator("cuda").manual_seed(1) for _ in range(len(prompts))]
images = pipeline(prompt=prompts, generator=generator, num_inference_steps=25).images
image_grid(images)
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/stable_diffusion_101/sd_101_8.png">
</div>

## 다음 단계

이 튜토리얼에서는 계산 및 메모리 효율을 높이고 생성된 출력의 품질을 개선하기 위해 `DiffusionPipeline`을 최적화하는 방법을 배웠습니다. 파이프라인을 더 빠르게 만드는 데 관심이 있다면 다음 리소스를 살펴보세요:

- [PyTorch 2.0](./optimization/torch2.0) 및 [`torch.compile`](https://pytorch.org/docs/stable/generated/torch.compile.html)이 어떻게 추론 속도를 5~300% 향상시킬 수 있는지 알아보세요. A100 GPU에서는 추론 속도가 최대 50%까지 빨라질 수 있습니다!
- PyTorch 2를 사용할 수 없는 경우, [xFormers](./optimization/xformers)를 설치하는 것이 좋습니다. 메모리 효율적인 어텐션 메커니즘은 PyTorch 1.13.1과 함께 사용하면 속도가 빨라지고 메모리 소비가 줄어듭니다.
- 모델 오프로딩과 같은 다른 최적화 기법은 [이 가이드](./optimization/fp16)에서 다루고 있습니다.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/stable_diffusion.md" />

### Diffusers
https://huggingface.co/docs/diffusers/main/ko/index.md

# Diffusers

🤗 Diffusers는 이미지, 오디오, 심지어 분자의 3D 구조를 생성하기 위한 최첨단 사전 훈련된 diffusion 모델을 위한 라이브러리입니다. 간단한 추론 솔루션을 찾고 있든, 자체 diffusion 모델을 훈련하고 싶든, 🤗 Diffusers는 두 가지 모두를 지원하는 모듈식 툴박스입니다. 저희 라이브러리는 [성능보다 사용성](conceptual/philosophy#usability-over-performance), [간편함보다 단순함](conceptual/philosophy#simple-over-easy), 그리고 [추상화보다 사용자 지정 가능성](conceptual/philosophy#tweakable-contributorfriendly-over-abstraction)에 중점을 두고 설계되었습니다.

이 라이브러리에는 세 가지 주요 구성 요소가 있습니다:

- 몇 줄의 코드만으로 추론할 수 있는 최첨단 [diffusion 파이프라인](api/pipelines/overview).
- 생성 속도와 품질 간의 균형을 맞추기 위해 상호교환적으로 사용할 수 있는 [노이즈 스케줄러](api/schedulers/overview).
- 빌딩 블록으로 사용할 수 있고 스케줄러와 결합하여 자체적인 end-to-end diffusion 시스템을 만들 수 있는 사전 학습된 [모델](api/models).

<div class="mt-10">
  <div class="w-full flex flex-col space-y-4 md:space-y-0 md:grid md:grid-cols-2 md:gap-y-4 md:gap-x-5">
    <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="./tutorials/tutorial_overview"
      ><div class="w-full text-center bg-gradient-to-br from-blue-400 to-blue-500 rounded-lg py-1.5 font-semibold mb-5 text-white text-lg leading-relaxed">Tutorials</div>
      <p class="text-gray-700">결과물을 생성하고, 나만의 diffusion 시스템을 구축하고, 확산 모델을 훈련하는 데 필요한 기본 기술을 배워보세요. 🤗 Diffusers를 처음 사용하는 경우 여기에서 시작하는 것이 좋습니다!</p>
    </a>
    <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="./using-diffusers/loading_overview"
      ><div class="w-full text-center bg-gradient-to-br from-indigo-400 to-indigo-500 rounded-lg py-1.5 font-semibold mb-5 text-white text-lg leading-relaxed">How-to guides</div>
      <p class="text-gray-700">파이프라인, 모델, 스케줄러를 로드하는 데 도움이 되는 실용적인 가이드입니다. 또한 특정 작업에 파이프라인을 사용하고, 출력 생성 방식을 제어하고, 추론 속도에 맞게 최적화하고, 다양한 학습 기법을 사용하는 방법도 배울 수 있습니다.</p>
    </a>
    <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="./conceptual/philosophy"
      ><div class="w-full text-center bg-gradient-to-br from-pink-400 to-pink-500 rounded-lg py-1.5 font-semibold mb-5 text-white text-lg leading-relaxed">Conceptual guides</div>
      <p class="text-gray-700">라이브러리가 왜 이런 방식으로 설계되었는지 이해하고, 라이브러리 이용에 대한 윤리적 가이드라인과 안전 구현에 대해 자세히 알아보세요.</p>
   </a>
    <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="./api/models"
      ><div class="w-full text-center bg-gradient-to-br from-purple-400 to-purple-500 rounded-lg py-1.5 font-semibold mb-5 text-white text-lg leading-relaxed">Reference</div>
      <p class="text-gray-700">🤗 Diffusers 클래스 및 메서드의 작동 방식에 대한 기술 설명.</p>
    </a>
  </div>
</div>

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/index.md" />

### Stable diffusion XL
https://huggingface.co/docs/diffusers/main/ko/api/pipelines/stable_diffusion/stable_diffusion_xl.md

# Stable diffusion XL

Stable Diffusion XL은 Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, Robin Rombach에 의해 [SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis](https://huggingface.co/papers/2307.01952)에서 제안되었습니다.

논문 초록은 다음을 따릅니다:

*text-to-image의 latent diffusion 모델인 SDXL을 소개합니다. 이전 버전의 Stable Diffusion과 비교하면, SDXL은 세 배 더큰 규모의 UNet 백본을 포함합니다: 모델 파라미터의 증가는 많은 attention 블럭을 사용하고 더 큰 cross-attention context를 SDXL의 두 번째 텍스트 인코더에 사용하기 때문입니다. 다중 종횡비에 다수의 새로운 conditioning 방법을 구성했습니다. 또한 후에 수정하는 image-to-image 기술을 사용함으로써 SDXL에 의해 생성된 시각적 품질을 향상하기 위해 정제된 모델을 소개합니다. SDXL은 이전 버전의 Stable Diffusion보다 성능이 향상되었고, 이러한 black-box 최신 이미지 생성자와 경쟁력있는 결과를 달성했습니다.*

## 팁

- Stable Diffusion XL은 특히 786과 1024사이의 이미지에 잘 작동합니다.
- Stable Diffusion XL은 아래와 같이 학습된 각 텍스트 인코더에 대해 서로 다른 프롬프트를 전달할 수 있습니다. 동일한 프롬프트의 다른 부분을 텍스트 인코더에 전달할 수도 있습니다.
- Stable Diffusion XL 결과 이미지는 아래에 보여지듯이 정제기(refiner)를 사용함으로써 향상될 수 있습니다.

### 이용가능한 체크포인트:

- *Text-to-Image (1024x1024 해상도)*: [StableDiffusionXLPipeline](/docs/diffusers/main/ko/api/pipelines/stable_diffusion/stable_diffusion_xl#diffusers.StableDiffusionXLPipeline)을 사용한 [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
- *Image-to-Image / 정제기(refiner) (1024x1024 해상도)*: [StableDiffusionXLImg2ImgPipeline](/docs/diffusers/main/ko/api/pipelines/stable_diffusion/stable_diffusion_xl#diffusers.StableDiffusionXLImg2ImgPipeline)를 사용한 [stabilityai/stable-diffusion-xl-refiner-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0)

## 사용 예시

SDXL을 사용하기 전에 `transformers`, `accelerate`, `safetensors` 와 `invisible_watermark`를 설치하세요.
다음과 같이 라이브러리를 설치할 수 있습니다:

```sh
pip install transformers
pip install accelerate
pip install safetensors
pip install invisible-watermark>=0.2.0
```

### 워터마커

Stable Diffusion XL로 이미지를 생성할 때 워터마크가 보이지 않도록 추가하는 것을 권장하는데, 이는 다운스트림(downstream) 어플리케이션에서 기계에 합성되었는지를 식별하는데 도움을 줄 수 있습니다. 그렇게 하려면 [invisible_watermark 라이브러리](https://pypi.org/project/invisible-watermark/)를 통해 설치해주세요:


```sh
pip install invisible-watermark>=0.2.0
```

`invisible-watermark` 라이브러리가 설치되면 워터마커가 **기본적으로** 사용될 것입니다.

생성 또는 안전하게 이미지를 배포하기 위해 다른 규정이 있다면, 다음과 같이 워터마커를 비활성화할 수 있습니다:

```py
pipe = StableDiffusionXLPipeline.from_pretrained(..., add_watermarker=False)
```

### Text-to-Image

*text-to-image*를 위해 다음과 같이 SDXL을 사용할 수 있습니다:

```py
from diffusers import StableDiffusionXLPipeline
import torch

pipe = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
)
pipe.to("cuda")

prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
image = pipe(prompt=prompt).images[0]
```

### Image-to-image

*image-to-image*를 위해 다음과 같이 SDXL을 사용할 수 있습니다:

```py
import torch
from diffusers import StableDiffusionXLImg2ImgPipeline
from diffusers.utils import load_image

pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
)
pipe = pipe.to("cuda")
url = "https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/aa_xl/000000009.png"

init_image = load_image(url).convert("RGB")
prompt = "a photo of an astronaut riding a horse on mars"
image = pipe(prompt, image=init_image).images[0]
```

### 인페인팅

*inpainting*를 위해 다음과 같이 SDXL을 사용할 수 있습니다:

```py
import torch
from diffusers import StableDiffusionXLInpaintPipeline
from diffusers.utils import load_image

pipe = StableDiffusionXLInpaintPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
)
pipe.to("cuda")

img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"

init_image = load_image(img_url).convert("RGB")
mask_image = load_image(mask_url).convert("RGB")

prompt = "A majestic tiger sitting on a bench"
image = pipe(prompt=prompt, image=init_image, mask_image=mask_image, num_inference_steps=50, strength=0.80).images[0]
```

### 이미지 결과물을 정제하기

[base 모델 체크포인트](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)에서, StableDiffusion-XL 또한 고주파 품질을 향상시키는 이미지를 생성하기 위해 낮은 노이즈 단계 이미지를 제거하는데 특화된 [refiner 체크포인트](https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0)를 포함하고 있습니다. 이 refiner 체크포인트는 이미지 품질을 향상시키기 위해 base 체크포인트를 실행한 후 "두 번째 단계" 파이프라인에 사용될 수 있습니다.

refiner를 사용할 때, 쉽게 사용할 수 있습니다
- 1.) base 모델과 refiner을 사용하는데, 이는 *Denoisers의 앙상블*을 위한 첫 번째 제안된 [eDiff-I](https://research.nvidia.com/labs/dir/eDiff-I/)를 사용하거나
- 2.) base 모델을 거친 후 [SDEdit](https://huggingface.co/papers/2108.01073) 방법으로 단순하게 refiner를 실행시킬 수 있습니다.

**참고**: SD-XL base와 refiner를 앙상블로 사용하는 아이디어는 커뮤니티 기여자들이 처음으로 제안했으며, 이는 다음과 같은 `diffusers`를 구현하는 데도 도움을 주셨습니다.
- [SytanSD](https://github.com/SytanSD)
- [bghira](https://github.com/bghira)
- [Birch-san](https://github.com/Birch-san)
- [AmericanPresidentJimmyCarter](https://github.com/AmericanPresidentJimmyCarter)

#### 1.) Denoisers의 앙상블

base와 refiner 모델을 denoiser의 앙상블로 사용할 때, base 모델은 고주파 diffusion 단계를 위한 전문가의 역할을 해야하고, refiner는 낮은 노이즈 diffusion 단계를 위한 전문가의 역할을 해야 합니다.

2.)에 비해 1.)의 장점은 전체적으로 denoising 단계가 덜 필요하므로 속도가 훨씬 더 빨라집니다. 단점은 base 모델의 결과를 검사할 수 없다는 것입니다. 즉, 여전히 노이즈가 심하게 제거됩니다.

base 모델과 refiner를 denoiser의 앙상블로 사용하기 위해 각각 고노이즈(high-nosise) (*즉* base 모델)와 저노이즈 (*즉* refiner 모델)의 노이즈를 제거하는 단계를 거쳐야하는 타임스텝의 기간을 정의해야 합니다.
base 모델의 [`denoising_end`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/stable_diffusion_xl#diffusers.StableDiffusionXLPipeline.__call__.denoising_end)와 refiner 모델의 [`denoising_start`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/stable_diffusion_xl#diffusers.StableDiffusionXLImg2ImgPipeline.__call__.denoising_start)를 사용해 간격을 정합니다.

`denoising_end`와 `denoising_start` 모두 0과 1사이의 실수 값으로 전달되어야 합니다.
전달되면 노이즈 제거의 끝과 시작은 모델 스케줄에 의해 정의된 이산적(discrete) 시간 간격의 비율로 정의됩니다.
노이즈 제거 단계의 수는 모델이 학습된 불연속적인 시간 간격과 선언된 fractional cutoff에 의해 결정되므로 '강도' 또한 선언된 경우 이 값이 '강도'를 재정의합니다.

예시를 들어보겠습니다.
우선, 두 개의 파이프라인을 가져옵니다. 텍스트 인코더와 variational autoencoder는 동일하므로 refiner를 위해 다시 불러오지 않아도 됩니다.

```py
from diffusers import DiffusionPipeline
import torch

base = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
)
pipe.to("cuda")

refiner = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-refiner-1.0",
    text_encoder_2=base.text_encoder_2,
    vae=base.vae,
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16",
)
refiner.to("cuda")
```

이제 추론 단계의 수와 고노이즈에서 노이즈를 제거하는 단계(*즉* base 모델)를 거쳐 실행되는 지점을 정의합니다.

```py
n_steps = 40
high_noise_frac = 0.8
```

Stable Diffusion XL base 모델은 타임스텝 0-999에 학습되며 Stable Diffusion XL refiner는 포괄적인 낮은 노이즈 타임스텝인 0-199에 base 모델로 부터 파인튜닝되어, 첫 800 타임스텝 (높은 노이즈)에 base 모델을 사용하고 마지막 200 타입스텝 (낮은 노이즈)에서 refiner가 사용됩니다. 따라서, `high_noise_frac`는 0.8로 설정하고, 모든 200-999 스텝(노이즈 제거 타임스텝의 첫 80%)은 base 모델에 의해 수행되며 0-199 스텝(노이즈 제거 타임스텝의 마지막 20%)은 refiner 모델에 의해 수행됩니다.

기억하세요, 노이즈 제거 절차는 **높은 값**(높은 노이즈) 타임스텝에서 시작되고, **낮은 값** (낮은 노이즈) 타임스텝에서 끝납니다.

이제 두 파이프라인을 실행해봅시다. `denoising_end`과 `denoising_start`를 같은 값으로 설정하고 `num_inference_steps`는 상수로 유지합니다. 또한 base 모델의 출력은 잠재 공간에 있어야 한다는 점을 기억하세요:

```py
prompt = "A majestic lion jumping from a big stone at night"

image = base(
    prompt=prompt,
    num_inference_steps=n_steps,
    denoising_end=high_noise_frac,
    output_type="latent",
).images
image = refiner(
    prompt=prompt,
    num_inference_steps=n_steps,
    denoising_start=high_noise_frac,
    image=image,
).images[0]
```

이미지를 살펴보겠습니다.

| 원래의 이미지 | Denoiser들의 앙상블 |
|---|---|
| ![lion_base](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lion_base.png) | ![lion_ref](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lion_refined.png)

동일한 40 단계에서 base 모델을 실행한다면, 이미지의 디테일(예: 사자의 눈과 코)이 떨어졌을 것입니다:

> [!TIP]
> 앙상블 방식은 사용 가능한 모든 스케줄러에서 잘 작동합니다!

#### 2.) 노이즈가 완전히 제거된 기본 이미지에서 이미지 출력을 정제하기

일반적인 `StableDiffusionImg2ImgPipeline` 방식에서, 기본 모델에서 생성된 완전히 노이즈가 제거된 이미지는 [refiner checkpoint](https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0)를 사용해 더 향상시킬 수 있습니다.

이를 위해, 보통의 "base" text-to-image 파이프라인을 수행 후에 image-to-image 파이프라인으로써 refiner를 실행시킬 수 있습니다. base 모델의 출력을 잠재 공간에 남겨둘 수 있습니다.

```py
from diffusers import DiffusionPipeline
import torch

pipe = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
)
pipe.to("cuda")

refiner = DiffusionPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-refiner-1.0",
    text_encoder_2=pipe.text_encoder_2,
    vae=pipe.vae,
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16",
)
refiner.to("cuda")

prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"

image = pipe(prompt=prompt, output_type="latent" if use_refiner else "pil").images[0]
image = refiner(prompt=prompt, image=image[None, :]).images[0]
```

| 원래의 이미지 | 정제된 이미지 |
|---|---|
| ![](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/sd_xl/init_image.png) | ![](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/sd_xl/refined_image.png) |

> [!TIP]
> refiner는 또한 인페인팅 설정에 잘 사용될 수 있습니다. 아래에 보여지듯이 [StableDiffusionXLInpaintPipeline](/docs/diffusers/main/ko/api/pipelines/stable_diffusion/stable_diffusion_xl#diffusers.StableDiffusionXLInpaintPipeline) 클래스를 사용해서 만들어보세요.

Denoiser 앙상블 설정에서 인페인팅에 refiner를 사용하려면 다음을 수행하면 됩니다:

```py
from diffusers import StableDiffusionXLInpaintPipeline
from diffusers.utils import load_image

pipe = StableDiffusionXLInpaintPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
)
pipe.to("cuda")

refiner = StableDiffusionXLInpaintPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-refiner-1.0",
    text_encoder_2=pipe.text_encoder_2,
    vae=pipe.vae,
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16",
)
refiner.to("cuda")

img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"

init_image = load_image(img_url).convert("RGB")
mask_image = load_image(mask_url).convert("RGB")

prompt = "A majestic tiger sitting on a bench"
num_inference_steps = 75
high_noise_frac = 0.7

image = pipe(
    prompt=prompt,
    image=init_image,
    mask_image=mask_image,
    num_inference_steps=num_inference_steps,
    denoising_start=high_noise_frac,
    output_type="latent",
).images
image = refiner(
    prompt=prompt,
    image=image,
    mask_image=mask_image,
    num_inference_steps=num_inference_steps,
    denoising_start=high_noise_frac,
).images[0]
```

일반적인 SDE 설정에서 인페인팅에 refiner를 사용하기 위해, `denoising_end`와 `denoising_start`를 제거하고 refiner의 추론 단계의 수를 적게 선택하세요.

### 단독 체크포인트 파일 / 원래의 파일 형식으로 불러오기

`from_single_file()`를 사용함으로써 원래의 파일 형식을 `diffusers` 형식으로 불러올 수 있습니다:

```py
from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
import torch

pipe = StableDiffusionXLPipeline.from_single_file(
    "./sd_xl_base_1.0.safetensors", torch_dtype=torch.float16
)
pipe.to("cuda")

refiner = StableDiffusionXLImg2ImgPipeline.from_single_file(
    "./sd_xl_refiner_1.0.safetensors", torch_dtype=torch.float16
)
refiner.to("cuda")
```

### 모델 offloading을 통해 메모리 최적화하기

out-of-memory 에러가 난다면, `StableDiffusionXLPipeline.enable_model_cpu_offload()`을 사용하는 것을 권장합니다.

```diff
- pipe.to("cuda")
+ pipe.enable_model_cpu_offload()
```

그리고

```diff
- refiner.to("cuda")
+ refiner.enable_model_cpu_offload()
```

### `torch.compile`로 추론 속도를 올리기

`torch.compile`를 사용함으로써 추론 속도를 올릴 수 있습니다. 이는 **ca.** 20% 속도 향상이 됩니다.

```diff
+ pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
+ refiner.unet = torch.compile(refiner.unet, mode="reduce-overhead", fullgraph=True)
```

### `torch < 2.0`일 때 실행하기

**참고** Stable Diffusion XL을 `torch`가 2.0 버전 미만에서 실행시키고 싶을 때, xformers 어텐션을 사용해주세요:

```sh
pip install xformers
```

```diff
+pipe.enable_xformers_memory_efficient_attention()
+refiner.enable_xformers_memory_efficient_attention()
```

## StableDiffusionXLPipeline[[diffusers.StableDiffusionXLPipeline]]

<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>class diffusers.StableDiffusionXLPipeline</name><anchor>diffusers.StableDiffusionXLPipeline</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py#L175</source><parameters>[{"name": "vae", "val": ": AutoencoderKL"}, {"name": "text_encoder", "val": ": CLIPTextModel"}, {"name": "text_encoder_2", "val": ": CLIPTextModelWithProjection"}, {"name": "tokenizer", "val": ": CLIPTokenizer"}, {"name": "tokenizer_2", "val": ": CLIPTokenizer"}, {"name": "unet", "val": ": UNet2DConditionModel"}, {"name": "scheduler", "val": ": KarrasDiffusionSchedulers"}, {"name": "image_encoder", "val": ": CLIPVisionModelWithProjection = None"}, {"name": "feature_extractor", "val": ": CLIPImageProcessor = None"}, {"name": "force_zeros_for_empty_prompt", "val": ": bool = True"}, {"name": "add_watermarker", "val": ": typing.Optional[bool] = None"}]</parameters><paramsdesc>- **vae** (`AutoencoderKL`) --
  Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
- **text_encoder** (`CLIPTextModel`) --
  Frozen text-encoder. Stable Diffusion XL uses the text portion of
  [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
  the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
- **text_encoder_2** (` CLIPTextModelWithProjection`) --
  Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
  [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
  specifically the
  [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
  variant.
- **tokenizer** (`CLIPTokenizer`) --
  Tokenizer of class
  [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
- **tokenizer_2** (`CLIPTokenizer`) --
  Second Tokenizer of class
  [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
- **unet** (`UNet2DConditionModel`) -- Conditional U-Net architecture to denoise the encoded image latents.
- **scheduler** (`SchedulerMixin`) --
  A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
  `DDIMScheduler`, `LMSDiscreteScheduler`, or `PNDMScheduler`.
- **force_zeros_for_empty_prompt** (`bool`, *optional*, defaults to `"True"`) --
  Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of
  `stabilityai/stable-diffusion-xl-base-1-0`.
- **add_watermarker** (`bool`, *optional*) --
  Whether to use the [invisible_watermark library](https://github.com/ShieldMnt/invisible-watermark/) to
  watermark output images. If not defined, it will default to True if the package is installed, otherwise no
  watermarker will be used.</paramsdesc><paramgroups>0</paramgroups></docstring>

Pipeline for text-to-image generation using Stable Diffusion XL.

This model inherits from `DiffusionPipeline`. Check the superclass documentation for the generic methods the
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)

The pipeline also inherits the following loading methods:
- `load_textual_inversion()` for loading textual inversion embeddings
- `from_single_file()` for loading `.ckpt` files
- `load_lora_weights()` for loading LoRA weights
- `save_lora_weights()` for saving LoRA weights
- `load_ip_adapter()` for loading IP Adapters





<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>__call__</name><anchor>diffusers.StableDiffusionXLPipeline.__call__</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py#L836</source><parameters>[{"name": "prompt", "val": ": typing.Union[str, typing.List[str]] = None"}, {"name": "prompt_2", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "height", "val": ": typing.Optional[int] = None"}, {"name": "width", "val": ": typing.Optional[int] = None"}, {"name": "num_inference_steps", "val": ": int = 50"}, {"name": "timesteps", "val": ": typing.List[int] = None"}, {"name": "sigmas", "val": ": typing.List[float] = None"}, {"name": "denoising_end", "val": ": typing.Optional[float] = None"}, {"name": "guidance_scale", "val": ": float = 5.0"}, {"name": "negative_prompt", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "negative_prompt_2", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "num_images_per_prompt", "val": ": typing.Optional[int] = 1"}, {"name": "eta", "val": ": float = 0.0"}, {"name": "generator", "val": ": typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None"}, {"name": "latents", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "ip_adapter_image", "val": ": typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor], NoneType] = None"}, {"name": "ip_adapter_image_embeds", "val": ": typing.Optional[typing.List[torch.Tensor]] = None"}, {"name": "output_type", "val": ": typing.Optional[str] = 'pil'"}, {"name": "return_dict", "val": ": bool = True"}, {"name": "cross_attention_kwargs", "val": ": typing.Optional[typing.Dict[str, typing.Any]] = None"}, {"name": "guidance_rescale", "val": ": float = 0.0"}, {"name": "original_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "crops_coords_top_left", "val": ": typing.Tuple[int, int] = (0, 0)"}, {"name": "target_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "negative_original_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "negative_crops_coords_top_left", "val": ": typing.Tuple[int, int] = (0, 0)"}, {"name": "negative_target_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "clip_skip", "val": ": typing.Optional[int] = None"}, {"name": "callback_on_step_end", "val": ": typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None"}, {"name": "callback_on_step_end_tensor_inputs", "val": ": typing.List[str] = ['latents']"}, {"name": "**kwargs", "val": ""}]</parameters><paramsdesc>- **prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
  instead.
- **prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
  used in both text-encoders
- **height** (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor) --
  The height in pixels of the generated image. This is set to 1024 by default for the best results.
  Anything below 512 pixels won't work well for
  [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
  and checkpoints that are not specifically fine-tuned on low resolutions.
- **width** (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor) --
  The width in pixels of the generated image. This is set to 1024 by default for the best results.
  Anything below 512 pixels won't work well for
  [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
  and checkpoints that are not specifically fine-tuned on low resolutions.
- **num_inference_steps** (`int`, *optional*, defaults to 50) --
  The number of denoising steps. More denoising steps usually lead to a higher quality image at the
  expense of slower inference.
- **timesteps** (`List[int]`, *optional*) --
  Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
  in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
  passed will be used. Must be in descending order.
- **sigmas** (`List[float]`, *optional*) --
  Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
  their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
  will be used.
- **denoising_end** (`float`, *optional*) --
  When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
  completed before it is intentionally prematurely terminated. As a result, the returned sample will
  still retain a substantial amount of noise as determined by the discrete timesteps selected by the
  scheduler. The denoising_end parameter should ideally be utilized when this pipeline forms a part of a
  "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
  Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output)
- **guidance_scale** (`float`, *optional*, defaults to 5.0) --
  Guidance scale as defined in [Classifier-Free Diffusion
  Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
  of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
  `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
  the text `prompt`, usually at the expense of lower image quality.
- **negative_prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
  less than `1`).
- **negative_prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
- **num_images_per_prompt** (`int`, *optional*, defaults to 1) --
  The number of images to generate per prompt.
- **eta** (`float`, *optional*, defaults to 0.0) --
  Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
  applies to `schedulers.DDIMScheduler`, will be ignored for others.
- **generator** (`torch.Generator` or `List[torch.Generator]`, *optional*) --
  One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
  to make generation deterministic.
- **latents** (`torch.Tensor`, *optional*) --
  Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
  generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
  tensor will be generated by sampling using the supplied random `generator`.
- **prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
  provided, text embeddings will be generated from `prompt` input argument.
- **negative_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
  argument.
- **pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
- **negative_pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
  input argument.
- **ip_adapter_image** -- (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
- **ip_adapter_image_embeds** (`List[torch.Tensor]`, *optional*) --
  Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
  IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
  contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
  provided, embeddings are computed from the `ip_adapter_image` input argument.
- **output_type** (`str`, *optional*, defaults to `"pil"`) --
  The output format of the generate image. Choose between
  [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
- **return_dict** (`bool`, *optional*, defaults to `True`) --
  Whether or not to return a `~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput` instead
  of a plain tuple.
- **cross_attention_kwargs** (`dict`, *optional*) --
  A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
  `self.processor` in
  [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
- **guidance_rescale** (`float`, *optional*, defaults to 0.0) --
  Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
  Flawed](https://huggingface.co/papers/2305.08891) `guidance_scale` is defined as `φ` in equation 16. of
  [Common Diffusion Noise Schedules and Sample Steps are
  Flawed](https://huggingface.co/papers/2305.08891). Guidance rescale factor should fix overexposure when
  using zero terminal SNR.
- **original_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
  `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
  explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **crops_coords_top_left** (`Tuple[int]`, *optional*, defaults to (0, 0)) --
  `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
  `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
  `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **target_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  For most cases, `target_size` should be set to the desired height and width of the generated image. If
  not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
  section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **negative_original_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  To negatively condition the generation process based on a specific image resolution. Part of SDXL's
  micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **negative_crops_coords_top_left** (`Tuple[int]`, *optional*, defaults to (0, 0)) --
  To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
  micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **negative_target_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  To negatively condition the generation process based on a target image resolution. It should be as same
  as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **callback_on_step_end** (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*) --
  A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
  each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
  DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
  list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
- **callback_on_step_end_tensor_inputs** (`List`, *optional*) --
  The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
  will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
  `._callback_tensor_inputs` attribute of your pipeline class.</paramsdesc><paramgroups>0</paramgroups><rettype>`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput` or `tuple`</rettype><retdesc>`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput` if `return_dict` is True, otherwise a
`tuple`. When returning a tuple, the first element is a list with the generated images.</retdesc></docstring>

Function invoked when calling the pipeline for generation.



<ExampleCodeBlock anchor="diffusers.StableDiffusionXLPipeline.__call__.example">

Examples:
```py
>>> import torch
>>> from diffusers import StableDiffusionXLPipeline

>>> pipe = StableDiffusionXLPipeline.from_pretrained(
...     "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16
... )
>>> pipe = pipe.to("cuda")

>>> prompt = "a photo of an astronaut riding a horse on mars"
>>> image = pipe(prompt).images[0]
```

</ExampleCodeBlock>







</div>
<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>encode_prompt</name><anchor>diffusers.StableDiffusionXLPipeline.encode_prompt</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py#L288</source><parameters>[{"name": "prompt", "val": ": str"}, {"name": "prompt_2", "val": ": typing.Optional[str] = None"}, {"name": "device", "val": ": typing.Optional[torch.device] = None"}, {"name": "num_images_per_prompt", "val": ": int = 1"}, {"name": "do_classifier_free_guidance", "val": ": bool = True"}, {"name": "negative_prompt", "val": ": typing.Optional[str] = None"}, {"name": "negative_prompt_2", "val": ": typing.Optional[str] = None"}, {"name": "prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "lora_scale", "val": ": typing.Optional[float] = None"}, {"name": "clip_skip", "val": ": typing.Optional[int] = None"}]</parameters><paramsdesc>- **prompt** (`str` or `List[str]`, *optional*) --
  prompt to be encoded
- **prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
  used in both text-encoders
- **device** -- (`torch.device`):
  torch device
- **num_images_per_prompt** (`int`) --
  number of images that should be generated per prompt
- **do_classifier_free_guidance** (`bool`) --
  whether to use classifier free guidance or not
- **negative_prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
  less than `1`).
- **negative_prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
- **prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
  provided, text embeddings will be generated from `prompt` input argument.
- **negative_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
  argument.
- **pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
- **negative_pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
  input argument.
- **lora_scale** (`float`, *optional*) --
  A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
- **clip_skip** (`int`, *optional*) --
  Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
  the output of the pre-final layer will be used for computing the prompt embeddings.</paramsdesc><paramgroups>0</paramgroups></docstring>

Encodes the prompt into text encoder hidden states.




</div>
<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>get_guidance_scale_embedding</name><anchor>diffusers.StableDiffusionXLPipeline.get_guidance_scale_embedding</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py#L771</source><parameters>[{"name": "w", "val": ": Tensor"}, {"name": "embedding_dim", "val": ": int = 512"}, {"name": "dtype", "val": ": dtype = torch.float32"}]</parameters><paramsdesc>- **w** (`torch.Tensor`) --
  Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
- **embedding_dim** (`int`, *optional*, defaults to 512) --
  Dimension of the embeddings to generate.
- **dtype** (`torch.dtype`, *optional*, defaults to `torch.float32`) --
  Data type of the generated embeddings.</paramsdesc><paramgroups>0</paramgroups><rettype>`torch.Tensor`</rettype><retdesc>Embedding vectors with shape `(len(w), embedding_dim)`.</retdesc></docstring>

See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298








</div></div>

## StableDiffusionXLImg2ImgPipeline[[diffusers.StableDiffusionXLImg2ImgPipeline]]

<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>class diffusers.StableDiffusionXLImg2ImgPipeline</name><anchor>diffusers.StableDiffusionXLImg2ImgPipeline</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py#L192</source><parameters>[{"name": "vae", "val": ": AutoencoderKL"}, {"name": "text_encoder", "val": ": CLIPTextModel"}, {"name": "text_encoder_2", "val": ": CLIPTextModelWithProjection"}, {"name": "tokenizer", "val": ": CLIPTokenizer"}, {"name": "tokenizer_2", "val": ": CLIPTokenizer"}, {"name": "unet", "val": ": UNet2DConditionModel"}, {"name": "scheduler", "val": ": KarrasDiffusionSchedulers"}, {"name": "image_encoder", "val": ": CLIPVisionModelWithProjection = None"}, {"name": "feature_extractor", "val": ": CLIPImageProcessor = None"}, {"name": "requires_aesthetics_score", "val": ": bool = False"}, {"name": "force_zeros_for_empty_prompt", "val": ": bool = True"}, {"name": "add_watermarker", "val": ": typing.Optional[bool] = None"}]</parameters><paramsdesc>- **vae** (`AutoencoderKL`) --
  Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
- **text_encoder** (`CLIPTextModel`) --
  Frozen text-encoder. Stable Diffusion XL uses the text portion of
  [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
  the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
- **text_encoder_2** (` CLIPTextModelWithProjection`) --
  Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
  [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
  specifically the
  [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
  variant.
- **tokenizer** (`CLIPTokenizer`) --
  Tokenizer of class
  [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
- **tokenizer_2** (`CLIPTokenizer`) --
  Second Tokenizer of class
  [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
- **unet** (`UNet2DConditionModel`) -- Conditional U-Net architecture to denoise the encoded image latents.
- **scheduler** (`SchedulerMixin`) --
  A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
  `DDIMScheduler`, `LMSDiscreteScheduler`, or `PNDMScheduler`.
- **requires_aesthetics_score** (`bool`, *optional*, defaults to `"False"`) --
  Whether the `unet` requires an `aesthetic_score` condition to be passed during inference. Also see the
  config of `stabilityai/stable-diffusion-xl-refiner-1-0`.
- **force_zeros_for_empty_prompt** (`bool`, *optional*, defaults to `"True"`) --
  Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of
  `stabilityai/stable-diffusion-xl-base-1-0`.
- **add_watermarker** (`bool`, *optional*) --
  Whether to use the [invisible_watermark library](https://github.com/ShieldMnt/invisible-watermark/) to
  watermark output images. If not defined, it will default to True if the package is installed, otherwise no
  watermarker will be used.</paramsdesc><paramgroups>0</paramgroups></docstring>

Pipeline for text-to-image generation using Stable Diffusion XL.

This model inherits from `DiffusionPipeline`. Check the superclass documentation for the generic methods the
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)

The pipeline also inherits the following loading methods:
- `load_textual_inversion()` for loading textual inversion embeddings
- `from_single_file()` for loading `.ckpt` files
- `load_lora_weights()` for loading LoRA weights
- `save_lora_weights()` for saving LoRA weights
- `load_ip_adapter()` for loading IP Adapters





<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>__call__</name><anchor>diffusers.StableDiffusionXLImg2ImgPipeline.__call__</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py#L986</source><parameters>[{"name": "prompt", "val": ": typing.Union[str, typing.List[str]] = None"}, {"name": "prompt_2", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "image", "val": ": typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor]] = None"}, {"name": "strength", "val": ": float = 0.3"}, {"name": "num_inference_steps", "val": ": int = 50"}, {"name": "timesteps", "val": ": typing.List[int] = None"}, {"name": "sigmas", "val": ": typing.List[float] = None"}, {"name": "denoising_start", "val": ": typing.Optional[float] = None"}, {"name": "denoising_end", "val": ": typing.Optional[float] = None"}, {"name": "guidance_scale", "val": ": float = 5.0"}, {"name": "negative_prompt", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "negative_prompt_2", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "num_images_per_prompt", "val": ": typing.Optional[int] = 1"}, {"name": "eta", "val": ": float = 0.0"}, {"name": "generator", "val": ": typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None"}, {"name": "latents", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "ip_adapter_image", "val": ": typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor], NoneType] = None"}, {"name": "ip_adapter_image_embeds", "val": ": typing.Optional[typing.List[torch.Tensor]] = None"}, {"name": "output_type", "val": ": typing.Optional[str] = 'pil'"}, {"name": "return_dict", "val": ": bool = True"}, {"name": "cross_attention_kwargs", "val": ": typing.Optional[typing.Dict[str, typing.Any]] = None"}, {"name": "guidance_rescale", "val": ": float = 0.0"}, {"name": "original_size", "val": ": typing.Tuple[int, int] = None"}, {"name": "crops_coords_top_left", "val": ": typing.Tuple[int, int] = (0, 0)"}, {"name": "target_size", "val": ": typing.Tuple[int, int] = None"}, {"name": "negative_original_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "negative_crops_coords_top_left", "val": ": typing.Tuple[int, int] = (0, 0)"}, {"name": "negative_target_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "aesthetic_score", "val": ": float = 6.0"}, {"name": "negative_aesthetic_score", "val": ": float = 2.5"}, {"name": "clip_skip", "val": ": typing.Optional[int] = None"}, {"name": "callback_on_step_end", "val": ": typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None"}, {"name": "callback_on_step_end_tensor_inputs", "val": ": typing.List[str] = ['latents']"}, {"name": "**kwargs", "val": ""}]</parameters><paramsdesc>- **prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
  instead.
- **prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
  used in both text-encoders
- **image** (`torch.Tensor` or `PIL.Image.Image` or `np.ndarray` or `List[torch.Tensor]` or `List[PIL.Image.Image]` or `List[np.ndarray]`) --
  The image(s) to modify with the pipeline.
- **strength** (`float`, *optional*, defaults to 0.3) --
  Conceptually, indicates how much to transform the reference `image`. Must be between 0 and 1. `image`
  will be used as a starting point, adding more noise to it the larger the `strength`. The number of
  denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise will
  be maximum and the denoising process will run for the full number of iterations specified in
  `num_inference_steps`. A value of 1, therefore, essentially ignores `image`. Note that in the case of
  `denoising_start` being declared as an integer, the value of `strength` will be ignored.
- **num_inference_steps** (`int`, *optional*, defaults to 50) --
  The number of denoising steps. More denoising steps usually lead to a higher quality image at the
  expense of slower inference.
- **timesteps** (`List[int]`, *optional*) --
  Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
  in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
  passed will be used. Must be in descending order.
- **sigmas** (`List[float]`, *optional*) --
  Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
  their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
  will be used.
- **denoising_start** (`float`, *optional*) --
  When specified, indicates the fraction (between 0.0 and 1.0) of the total denoising process to be
  bypassed before it is initiated. Consequently, the initial part of the denoising process is skipped and
  it is assumed that the passed `image` is a partly denoised image. Note that when this is specified,
  strength will be ignored. The `denoising_start` parameter is particularly beneficial when this pipeline
  is integrated into a "Mixture of Denoisers" multi-pipeline setup, as detailed in [**Refine Image
  Quality**](https://huggingface.co/docs/diffusers/using-diffusers/sdxl#refine-image-quality).
- **denoising_end** (`float`, *optional*) --
  When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
  completed before it is intentionally prematurely terminated. As a result, the returned sample will
  still retain a substantial amount of noise (ca. final 20% of timesteps still needed) and should be
  denoised by a successor pipeline that has `denoising_start` set to 0.8 so that it only denoises the
  final 20% of the scheduler. The denoising_end parameter should ideally be utilized when this pipeline
  forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refine Image
  Quality**](https://huggingface.co/docs/diffusers/using-diffusers/sdxl#refine-image-quality).
- **guidance_scale** (`float`, *optional*, defaults to 7.5) --
  Guidance scale as defined in [Classifier-Free Diffusion
  Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
  of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
  `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
  the text `prompt`, usually at the expense of lower image quality.
- **negative_prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
  less than `1`).
- **negative_prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
- **num_images_per_prompt** (`int`, *optional*, defaults to 1) --
  The number of images to generate per prompt.
- **eta** (`float`, *optional*, defaults to 0.0) --
  Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
  applies to `schedulers.DDIMScheduler`, will be ignored for others.
- **generator** (`torch.Generator` or `List[torch.Generator]`, *optional*) --
  One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
  to make generation deterministic.
- **latents** (`torch.Tensor`, *optional*) --
  Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
  generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
  tensor will be generated by sampling using the supplied random `generator`.
- **prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
  provided, text embeddings will be generated from `prompt` input argument.
- **negative_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
  argument.
- **pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
- **negative_pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
  input argument.
- **ip_adapter_image** -- (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
- **ip_adapter_image_embeds** (`List[torch.Tensor]`, *optional*) --
  Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
  IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
  contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
  provided, embeddings are computed from the `ip_adapter_image` input argument.
- **output_type** (`str`, *optional*, defaults to `"pil"`) --
  The output format of the generate image. Choose between
  [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
- **return_dict** (`bool`, *optional*, defaults to `True`) --
  Whether or not to return a `~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput` instead of a
  plain tuple.
- **cross_attention_kwargs** (`dict`, *optional*) --
  A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
  `self.processor` in
  [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
- **guidance_rescale** (`float`, *optional*, defaults to 0.0) --
  Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
  Flawed](https://huggingface.co/papers/2305.08891) `guidance_scale` is defined as `φ` in equation 16. of
  [Common Diffusion Noise Schedules and Sample Steps are
  Flawed](https://huggingface.co/papers/2305.08891). Guidance rescale factor should fix overexposure when
  using zero terminal SNR.
- **original_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
  `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
  explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **crops_coords_top_left** (`Tuple[int]`, *optional*, defaults to (0, 0)) --
  `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
  `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
  `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **target_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  For most cases, `target_size` should be set to the desired height and width of the generated image. If
  not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
  section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **negative_original_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  To negatively condition the generation process based on a specific image resolution. Part of SDXL's
  micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **negative_crops_coords_top_left** (`Tuple[int]`, *optional*, defaults to (0, 0)) --
  To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
  micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **negative_target_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  To negatively condition the generation process based on a target image resolution. It should be as same
  as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **aesthetic_score** (`float`, *optional*, defaults to 6.0) --
  Used to simulate an aesthetic score of the generated image by influencing the positive text condition.
  Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **negative_aesthetic_score** (`float`, *optional*, defaults to 2.5) --
  Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
  simulate an aesthetic score of the generated image by influencing the negative text condition.
- **clip_skip** (`int`, *optional*) --
  Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
  the output of the pre-final layer will be used for computing the prompt embeddings.
- **callback_on_step_end** (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*) --
  A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
  each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
  DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
  list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
- **callback_on_step_end_tensor_inputs** (`List`, *optional*) --
  The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
  will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
  `._callback_tensor_inputs` attribute of your pipeline class.</paramsdesc><paramgroups>0</paramgroups><rettype>`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput` or `tuple`</rettype><retdesc>`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput` if `return_dict` is True, otherwise a
`tuple. When returning a tuple, the first element is a list with the generated images.</retdesc></docstring>

Function invoked when calling the pipeline for generation.



<ExampleCodeBlock anchor="diffusers.StableDiffusionXLImg2ImgPipeline.__call__.example">

Examples:
```py
>>> import torch
>>> from diffusers import StableDiffusionXLImg2ImgPipeline
>>> from diffusers.utils import load_image

>>> pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
...     "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16
... )
>>> pipe = pipe.to("cuda")
>>> url = "https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/aa_xl/000000009.png"

>>> init_image = load_image(url).convert("RGB")
>>> prompt = "a photo of an astronaut riding a horse on mars"
>>> image = pipe(prompt, image=init_image).images[0]
```

</ExampleCodeBlock>







</div>
<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>encode_prompt</name><anchor>diffusers.StableDiffusionXLImg2ImgPipeline.encode_prompt</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py#L305</source><parameters>[{"name": "prompt", "val": ": str"}, {"name": "prompt_2", "val": ": typing.Optional[str] = None"}, {"name": "device", "val": ": typing.Optional[torch.device] = None"}, {"name": "num_images_per_prompt", "val": ": int = 1"}, {"name": "do_classifier_free_guidance", "val": ": bool = True"}, {"name": "negative_prompt", "val": ": typing.Optional[str] = None"}, {"name": "negative_prompt_2", "val": ": typing.Optional[str] = None"}, {"name": "prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "lora_scale", "val": ": typing.Optional[float] = None"}, {"name": "clip_skip", "val": ": typing.Optional[int] = None"}]</parameters><paramsdesc>- **prompt** (`str` or `List[str]`, *optional*) --
  prompt to be encoded
- **prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
  used in both text-encoders
- **device** -- (`torch.device`):
  torch device
- **num_images_per_prompt** (`int`) --
  number of images that should be generated per prompt
- **do_classifier_free_guidance** (`bool`) --
  whether to use classifier free guidance or not
- **negative_prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
  less than `1`).
- **negative_prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
- **prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
  provided, text embeddings will be generated from `prompt` input argument.
- **negative_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
  argument.
- **pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
- **negative_pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
  input argument.
- **lora_scale** (`float`, *optional*) --
  A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
- **clip_skip** (`int`, *optional*) --
  Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
  the output of the pre-final layer will be used for computing the prompt embeddings.</paramsdesc><paramgroups>0</paramgroups></docstring>

Encodes the prompt into text encoder hidden states.




</div>
<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>get_guidance_scale_embedding</name><anchor>diffusers.StableDiffusionXLImg2ImgPipeline.get_guidance_scale_embedding</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py#L917</source><parameters>[{"name": "w", "val": ": Tensor"}, {"name": "embedding_dim", "val": ": int = 512"}, {"name": "dtype", "val": ": dtype = torch.float32"}]</parameters><paramsdesc>- **w** (`torch.Tensor`) --
  Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
- **embedding_dim** (`int`, *optional*, defaults to 512) --
  Dimension of the embeddings to generate.
- **dtype** (`torch.dtype`, *optional*, defaults to `torch.float32`) --
  Data type of the generated embeddings.</paramsdesc><paramgroups>0</paramgroups><rettype>`torch.Tensor`</rettype><retdesc>Embedding vectors with shape `(len(w), embedding_dim)`.</retdesc></docstring>

See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298








</div></div>

## StableDiffusionXLInpaintPipeline[[diffusers.StableDiffusionXLInpaintPipeline]]

<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>class diffusers.StableDiffusionXLInpaintPipeline</name><anchor>diffusers.StableDiffusionXLInpaintPipeline</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py#L219</source><parameters>[{"name": "vae", "val": ": AutoencoderKL"}, {"name": "text_encoder", "val": ": CLIPTextModel"}, {"name": "text_encoder_2", "val": ": CLIPTextModelWithProjection"}, {"name": "tokenizer", "val": ": CLIPTokenizer"}, {"name": "tokenizer_2", "val": ": CLIPTokenizer"}, {"name": "unet", "val": ": UNet2DConditionModel"}, {"name": "scheduler", "val": ": KarrasDiffusionSchedulers"}, {"name": "image_encoder", "val": ": CLIPVisionModelWithProjection = None"}, {"name": "feature_extractor", "val": ": CLIPImageProcessor = None"}, {"name": "requires_aesthetics_score", "val": ": bool = False"}, {"name": "force_zeros_for_empty_prompt", "val": ": bool = True"}, {"name": "add_watermarker", "val": ": typing.Optional[bool] = None"}]</parameters><paramsdesc>- **vae** (`AutoencoderKL`) --
  Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
- **text_encoder** (`CLIPTextModel`) --
  Frozen text-encoder. Stable Diffusion XL uses the text portion of
  [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
  the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
- **text_encoder_2** (` CLIPTextModelWithProjection`) --
  Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
  [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
  specifically the
  [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
  variant.
- **tokenizer** (`CLIPTokenizer`) --
  Tokenizer of class
  [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
- **tokenizer_2** (`CLIPTokenizer`) --
  Second Tokenizer of class
  [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
- **unet** (`UNet2DConditionModel`) -- Conditional U-Net architecture to denoise the encoded image latents.
- **scheduler** (`SchedulerMixin`) --
  A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
  `DDIMScheduler`, `LMSDiscreteScheduler`, or `PNDMScheduler`.
- **requires_aesthetics_score** (`bool`, *optional*, defaults to `"False"`) --
  Whether the `unet` requires a aesthetic_score condition to be passed during inference. Also see the config
  of `stabilityai/stable-diffusion-xl-refiner-1-0`.
- **force_zeros_for_empty_prompt** (`bool`, *optional*, defaults to `"True"`) --
  Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of
  `stabilityai/stable-diffusion-xl-base-1-0`.
- **add_watermarker** (`bool`, *optional*) --
  Whether to use the [invisible_watermark library](https://github.com/ShieldMnt/invisible-watermark/) to
  watermark output images. If not defined, it will default to True if the package is installed, otherwise no
  watermarker will be used.</paramsdesc><paramgroups>0</paramgroups></docstring>

Pipeline for text-to-image generation using Stable Diffusion XL.

This model inherits from `DiffusionPipeline`. Check the superclass documentation for the generic methods the
library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)

The pipeline also inherits the following loading methods:
- `load_textual_inversion()` for loading textual inversion embeddings
- `from_single_file()` for loading `.ckpt` files
- `load_lora_weights()` for loading LoRA weights
- `save_lora_weights()` for saving LoRA weights
- `load_ip_adapter()` for loading IP Adapters





<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>__call__</name><anchor>diffusers.StableDiffusionXLInpaintPipeline.__call__</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py#L1091</source><parameters>[{"name": "prompt", "val": ": typing.Union[str, typing.List[str]] = None"}, {"name": "prompt_2", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "image", "val": ": typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor]] = None"}, {"name": "mask_image", "val": ": typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor]] = None"}, {"name": "masked_image_latents", "val": ": Tensor = None"}, {"name": "height", "val": ": typing.Optional[int] = None"}, {"name": "width", "val": ": typing.Optional[int] = None"}, {"name": "padding_mask_crop", "val": ": typing.Optional[int] = None"}, {"name": "strength", "val": ": float = 0.9999"}, {"name": "num_inference_steps", "val": ": int = 50"}, {"name": "timesteps", "val": ": typing.List[int] = None"}, {"name": "sigmas", "val": ": typing.List[float] = None"}, {"name": "denoising_start", "val": ": typing.Optional[float] = None"}, {"name": "denoising_end", "val": ": typing.Optional[float] = None"}, {"name": "guidance_scale", "val": ": float = 7.5"}, {"name": "negative_prompt", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "negative_prompt_2", "val": ": typing.Union[str, typing.List[str], NoneType] = None"}, {"name": "num_images_per_prompt", "val": ": typing.Optional[int] = 1"}, {"name": "eta", "val": ": float = 0.0"}, {"name": "generator", "val": ": typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None"}, {"name": "latents", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "ip_adapter_image", "val": ": typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor], NoneType] = None"}, {"name": "ip_adapter_image_embeds", "val": ": typing.Optional[typing.List[torch.Tensor]] = None"}, {"name": "output_type", "val": ": typing.Optional[str] = 'pil'"}, {"name": "return_dict", "val": ": bool = True"}, {"name": "cross_attention_kwargs", "val": ": typing.Optional[typing.Dict[str, typing.Any]] = None"}, {"name": "guidance_rescale", "val": ": float = 0.0"}, {"name": "original_size", "val": ": typing.Tuple[int, int] = None"}, {"name": "crops_coords_top_left", "val": ": typing.Tuple[int, int] = (0, 0)"}, {"name": "target_size", "val": ": typing.Tuple[int, int] = None"}, {"name": "negative_original_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "negative_crops_coords_top_left", "val": ": typing.Tuple[int, int] = (0, 0)"}, {"name": "negative_target_size", "val": ": typing.Optional[typing.Tuple[int, int]] = None"}, {"name": "aesthetic_score", "val": ": float = 6.0"}, {"name": "negative_aesthetic_score", "val": ": float = 2.5"}, {"name": "clip_skip", "val": ": typing.Optional[int] = None"}, {"name": "callback_on_step_end", "val": ": typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None"}, {"name": "callback_on_step_end_tensor_inputs", "val": ": typing.List[str] = ['latents']"}, {"name": "**kwargs", "val": ""}]</parameters><paramsdesc>- **prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
  instead.
- **prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
  used in both text-encoders
- **image** (`PIL.Image.Image`) --
  `Image`, or tensor representing an image batch which will be inpainted, *i.e.* parts of the image will
  be masked out with `mask_image` and repainted according to `prompt`.
- **mask_image** (`PIL.Image.Image`) --
  `Image`, or tensor representing an image batch, to mask `image`. White pixels in the mask will be
  repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted
  to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L)
  instead of 3, so the expected shape would be `(B, H, W, 1)`.
- **height** (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor) --
  The height in pixels of the generated image. This is set to 1024 by default for the best results.
  Anything below 512 pixels won't work well for
  [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
  and checkpoints that are not specifically fine-tuned on low resolutions.
- **width** (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor) --
  The width in pixels of the generated image. This is set to 1024 by default for the best results.
  Anything below 512 pixels won't work well for
  [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
  and checkpoints that are not specifically fine-tuned on low resolutions.
- **padding_mask_crop** (`int`, *optional*, defaults to `None`) --
  The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to
  image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region
  with the same aspect ration of the image and contains all masked area, and then expand that area based
  on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before
  resizing to the original image size for inpainting. This is useful when the masked area is small while
  the image is large and contain information irrelevant for inpainting, such as background.
- **strength** (`float`, *optional*, defaults to 0.9999) --
  Conceptually, indicates how much to transform the masked portion of the reference `image`. Must be
  between 0 and 1. `image` will be used as a starting point, adding more noise to it the larger the
  `strength`. The number of denoising steps depends on the amount of noise initially added. When
  `strength` is 1, added noise will be maximum and the denoising process will run for the full number of
  iterations specified in `num_inference_steps`. A value of 1, therefore, essentially ignores the masked
  portion of the reference `image`. Note that in the case of `denoising_start` being declared as an
  integer, the value of `strength` will be ignored.
- **num_inference_steps** (`int`, *optional*, defaults to 50) --
  The number of denoising steps. More denoising steps usually lead to a higher quality image at the
  expense of slower inference.
- **timesteps** (`List[int]`, *optional*) --
  Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
  in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
  passed will be used. Must be in descending order.
- **sigmas** (`List[float]`, *optional*) --
  Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
  their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
  will be used.
- **denoising_start** (`float`, *optional*) --
  When specified, indicates the fraction (between 0.0 and 1.0) of the total denoising process to be
  bypassed before it is initiated. Consequently, the initial part of the denoising process is skipped and
  it is assumed that the passed `image` is a partly denoised image. Note that when this is specified,
  strength will be ignored. The `denoising_start` parameter is particularly beneficial when this pipeline
  is integrated into a "Mixture of Denoisers" multi-pipeline setup, as detailed in [**Refining the Image
  Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
- **denoising_end** (`float`, *optional*) --
  When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
  completed before it is intentionally prematurely terminated. As a result, the returned sample will
  still retain a substantial amount of noise (ca. final 20% of timesteps still needed) and should be
  denoised by a successor pipeline that has `denoising_start` set to 0.8 so that it only denoises the
  final 20% of the scheduler. The denoising_end parameter should ideally be utilized when this pipeline
  forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
  Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
- **guidance_scale** (`float`, *optional*, defaults to 7.5) --
  Guidance scale as defined in [Classifier-Free Diffusion
  Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
  of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
  `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
  the text `prompt`, usually at the expense of lower image quality.
- **negative_prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
  less than `1`).
- **negative_prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
- **prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
  provided, text embeddings will be generated from `prompt` input argument.
- **negative_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
  argument.
- **pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
- **negative_pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
  input argument.
- **ip_adapter_image** -- (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
- **ip_adapter_image_embeds** (`List[torch.Tensor]`, *optional*) --
  Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
  IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
  contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
  provided, embeddings are computed from the `ip_adapter_image` input argument.
- **num_images_per_prompt** (`int`, *optional*, defaults to 1) --
  The number of images to generate per prompt.
- **eta** (`float`, *optional*, defaults to 0.0) --
  Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
  applies to `schedulers.DDIMScheduler`, will be ignored for others.
- **generator** (`torch.Generator`, *optional*) --
  One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
  to make generation deterministic.
- **latents** (`torch.Tensor`, *optional*) --
  Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
  generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
  tensor will be generated by sampling using the supplied random `generator`.
- **output_type** (`str`, *optional*, defaults to `"pil"`) --
  The output format of the generate image. Choose between
  [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
- **return_dict** (`bool`, *optional*, defaults to `True`) --
  Whether or not to return a `StableDiffusionPipelineOutput` instead of a
  plain tuple.
- **cross_attention_kwargs** (`dict`, *optional*) --
  A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
  `self.processor` in
  [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
- **original_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
  `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
  explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **crops_coords_top_left** (`Tuple[int]`, *optional*, defaults to (0, 0)) --
  `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
  `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
  `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **target_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  For most cases, `target_size` should be set to the desired height and width of the generated image. If
  not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
  section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **negative_original_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  To negatively condition the generation process based on a specific image resolution. Part of SDXL's
  micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **negative_crops_coords_top_left** (`Tuple[int]`, *optional*, defaults to (0, 0)) --
  To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
  micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **negative_target_size** (`Tuple[int]`, *optional*, defaults to (1024, 1024)) --
  To negatively condition the generation process based on a target image resolution. It should be as same
  as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
  information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
- **aesthetic_score** (`float`, *optional*, defaults to 6.0) --
  Used to simulate an aesthetic score of the generated image by influencing the positive text condition.
  Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
- **negative_aesthetic_score** (`float`, *optional*, defaults to 2.5) --
  Part of SDXL's micro-conditioning as explained in section 2.2 of
  [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
  simulate an aesthetic score of the generated image by influencing the negative text condition.
- **clip_skip** (`int`, *optional*) --
  Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
  the output of the pre-final layer will be used for computing the prompt embeddings.
- **callback_on_step_end** (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*) --
  A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
  each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
  DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
  list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
- **callback_on_step_end_tensor_inputs** (`List`, *optional*) --
  The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
  will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
  `._callback_tensor_inputs` attribute of your pipeline class.</paramsdesc><paramgroups>0</paramgroups><rettype>`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput` or `tuple`</rettype><retdesc>`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput` if `return_dict` is True, otherwise a
`tuple. `tuple. When returning a tuple, the first element is a list with the generated images.</retdesc></docstring>

Function invoked when calling the pipeline for generation.



<ExampleCodeBlock anchor="diffusers.StableDiffusionXLInpaintPipeline.__call__.example">

Examples:
```py
>>> import torch
>>> from diffusers import StableDiffusionXLInpaintPipeline
>>> from diffusers.utils import load_image

>>> pipe = StableDiffusionXLInpaintPipeline.from_pretrained(
...     "stabilityai/stable-diffusion-xl-base-1.0",
...     torch_dtype=torch.float16,
...     variant="fp16",
...     use_safetensors=True,
... )
>>> pipe.to("cuda")

>>> img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
>>> mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"

>>> init_image = load_image(img_url).convert("RGB")
>>> mask_image = load_image(mask_url).convert("RGB")

>>> prompt = "A majestic tiger sitting on a bench"
>>> image = pipe(
...     prompt=prompt, image=init_image, mask_image=mask_image, num_inference_steps=50, strength=0.80
... ).images[0]
```

</ExampleCodeBlock>







</div>
<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>encode_prompt</name><anchor>diffusers.StableDiffusionXLInpaintPipeline.encode_prompt</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py#L409</source><parameters>[{"name": "prompt", "val": ": str"}, {"name": "prompt_2", "val": ": typing.Optional[str] = None"}, {"name": "device", "val": ": typing.Optional[torch.device] = None"}, {"name": "num_images_per_prompt", "val": ": int = 1"}, {"name": "do_classifier_free_guidance", "val": ": bool = True"}, {"name": "negative_prompt", "val": ": typing.Optional[str] = None"}, {"name": "negative_prompt_2", "val": ": typing.Optional[str] = None"}, {"name": "prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "negative_pooled_prompt_embeds", "val": ": typing.Optional[torch.Tensor] = None"}, {"name": "lora_scale", "val": ": typing.Optional[float] = None"}, {"name": "clip_skip", "val": ": typing.Optional[int] = None"}]</parameters><paramsdesc>- **prompt** (`str` or `List[str]`, *optional*) --
  prompt to be encoded
- **prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
  used in both text-encoders
- **device** -- (`torch.device`):
  torch device
- **num_images_per_prompt** (`int`) --
  number of images that should be generated per prompt
- **do_classifier_free_guidance** (`bool`) --
  whether to use classifier free guidance or not
- **negative_prompt** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation. If not defined, one has to pass
  `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
  less than `1`).
- **negative_prompt_2** (`str` or `List[str]`, *optional*) --
  The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
  `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
- **prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
  provided, text embeddings will be generated from `prompt` input argument.
- **negative_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
  argument.
- **pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
  If not provided, pooled text embeddings will be generated from `prompt` input argument.
- **negative_pooled_prompt_embeds** (`torch.Tensor`, *optional*) --
  Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
  weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
  input argument.
- **lora_scale** (`float`, *optional*) --
  A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
- **clip_skip** (`int`, *optional*) --
  Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
  the output of the pre-final layer will be used for computing the prompt embeddings.</paramsdesc><paramgroups>0</paramgroups></docstring>

Encodes the prompt into text encoder hidden states.




</div>
<div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8">


<docstring><name>get_guidance_scale_embedding</name><anchor>diffusers.StableDiffusionXLInpaintPipeline.get_guidance_scale_embedding</anchor><source>https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py#L1022</source><parameters>[{"name": "w", "val": ": Tensor"}, {"name": "embedding_dim", "val": ": int = 512"}, {"name": "dtype", "val": ": dtype = torch.float32"}]</parameters><paramsdesc>- **w** (`torch.Tensor`) --
  Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
- **embedding_dim** (`int`, *optional*, defaults to 512) --
  Dimension of the embeddings to generate.
- **dtype** (`torch.dtype`, *optional*, defaults to `torch.float32`) --
  Data type of the generated embeddings.</paramsdesc><paramgroups>0</paramgroups><rettype>`torch.Tensor`</rettype><retdesc>Embedding vectors with shape `(len(w), embedding_dim)`.</retdesc></docstring>

See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298








</div></div>

### 각 텍스트 인코더에 다른 프롬프트를 전달하기

Stable Diffusion XL는 두 개의 텍스트 인코더에 학습되었습니다. 기본 동작은 각 프롬프트에 동일한 프롬프트를 전달하는 것입니다. 그러나 [일부 사용자](https://github.com/huggingface/diffusers/issues/4004#issuecomment-1627764201)가 품질을 향상시킬 수 있다고 지적한 것처럼 텍스트 인코더마다 다른 프롬프트를 전달할 수 있습니다. 그렇게 하려면, `prompt_2`와 `negative_prompt_2`를 `prompt`와 `negative_prompt`에 전달해야 합니다. 그렇게 함으로써, 원래의 프롬프트들(`prompt`)과 부정 프롬프트들(`negative_prompt`)를 `텍스트 인코더`에 전달할 것입니다.(공식 SDXL 0.9/1.0의 [OpenAI CLIP-ViT/L-14](https://huggingface.co/openai/clip-vit-large-patch14)에서 볼 수 있습니다.) 그리고 `prompt_2`와 `negative_prompt_2`는 `text_encoder_2`에 전달됩니다.(공식 SDXL 0.9/1.0의 [OpenCLIP-ViT/bigG-14](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)에서 볼 수 있습니다.)

```py
from diffusers import StableDiffusionXLPipeline
import torch

pipe = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-0.9", torch_dtype=torch.float16, variant="fp16", use_safetensors=True
)
pipe.to("cuda")

# OAI CLIP-ViT/L-14에 prompt가 전달됩니다
prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k"
# OpenCLIP-ViT/bigG-14에 prompt_2가 전달됩니다
prompt_2 = "monet painting"
image = pipe(prompt=prompt, prompt_2=prompt_2).images[0]
```

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/api/pipelines/stable_diffusion/stable_diffusion_xl.md" />

### DreamBooth
https://huggingface.co/docs/diffusers/main/ko/training/dreambooth.md

# DreamBooth

[DreamBooth](https://huggingface.co/papers/2208.12242)는 한 주제에 대한 적은 이미지(3~5개)만으로도 stable diffusion과 같이 text-to-image 모델을 개인화할 수 있는 방법입니다. 이를 통해 모델은 다양한 장면, 포즈 및 장면(뷰)에서 피사체에 대해 맥락화(contextualized)된 이미지를 생성할 수 있습니다.

![프로젝트 블로그에서의 DreamBooth 예시](https://dreambooth.github.io/DreamBooth_files/teaser_static.jpg)
<small>에서의 Dreambooth 예시 <a href="https://dreambooth.github.io">project's blog.</a></small>


이 가이드는 다양한 GPU, Flax 사양에 대해 [`CompVis/stable-diffusion-v1-4`](https://huggingface.co/CompVis/stable-diffusion-v1-4) 모델로 DreamBooth를 파인튜닝하는 방법을 보여줍니다. 더 깊이 파고들어 작동 방식을 확인하는 데 관심이 있는 경우, 이 가이드에 사용된 DreamBooth의 모든 학습 스크립트를 [여기](https://github.com/huggingface/diffusers/tree/main/examples/dreambooth)에서 찾을 수 있습니다.

스크립트를 실행하기 전에 라이브러리의 학습에 필요한 dependencies를 설치해야 합니다. 또한 `main` GitHub 브랜치에서 🧨 Diffusers를 설치하는 것이 좋습니다.

```bash
pip install git+https://github.com/huggingface/diffusers
pip install -U -r diffusers/examples/dreambooth/requirements.txt
```

xFormers는 학습에 필요한 요구 사항은 아니지만, 가능하면 [설치](../optimization/xformers)하는 것이 좋습니다. 학습 속도를 높이고 메모리 사용량을 줄일 수 있기 때문입니다.

모든 dependencies을 설정한 후 다음을 사용하여 [🤗 Accelerate](https://github.com/huggingface/accelerate/) 환경을 다음과 같이 초기화합니다:

```bash
accelerate config
```

별도 설정 없이 기본 🤗 Accelerate 환경을 설치하려면 다음을 실행합니다:

```bash
accelerate config default
```

또는 현재 환경이 노트북과 같은 대화형 셸을 지원하지 않는 경우 다음을 사용할 수 있습니다:

```py
from accelerate.utils import write_basic_config

write_basic_config()
```

## 파인튜닝

> [!WARNING]
> DreamBooth 파인튜닝은 하이퍼파라미터에 매우 민감하고 과적합되기 쉽습니다. 적절한 하이퍼파라미터를 선택하는 데 도움이 되도록 다양한 권장 설정이 포함된 [심층 분석](https://huggingface.co/blog/dreambooth)을 살펴보는 것이 좋습니다.

<frameworkcontent>
<pt>
[몇 장의 강아지 이미지들](https://drive.google.com/drive/folders/1BO_dyz-p65qhBRRMRA4TbZ8qW4rB99JZ)로 DreamBooth를 시도해봅시다.
이를 다운로드해 디렉터리에 저장한 다음 `INSTANCE_DIR` 환경 변수를 해당 경로로 설정합니다:


```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path_to_training_images"
export OUTPUT_DIR="path_to_saved_model"
```

그런 다음, 다음 명령을 사용하여 학습 스크립트를 실행할 수 있습니다 (전체 학습 스크립트는 [여기](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth.py)에서 찾을 수 있습니다):

```bash
accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --instance_prompt="a photo of sks dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=400
```
</pt>
<jax>

TPU에 액세스할 수 있거나 더 빠르게 훈련하고 싶다면 [Flax 학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth_flax.py)를 사용해 볼 수 있습니다. Flax 학습 스크립트는 gradient checkpointing 또는 gradient accumulation을 지원하지 않으므로, 메모리가 30GB 이상인 GPU가 필요합니다.

스크립트를 실행하기 전에 요구 사항이 설치되어 있는지 확인하십시오.

```bash
pip install -U -r requirements.txt
```

그러면 다음 명령어로 학습 스크립트를 실행시킬 수 있습니다:

```bash
export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
export INSTANCE_DIR="path-to-instance-images"
export OUTPUT_DIR="path-to-save-model"

python train_dreambooth_flax.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --instance_prompt="a photo of sks dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --learning_rate=5e-6 \
  --max_train_steps=400
```
</jax>
</frameworkcontent>

### Prior-preserving(사전 보존) loss를 사용한 파인튜닝

과적합과 language drift를 방지하기 위해 사전 보존이 사용됩니다(관심이 있는 경우 [논문](https://huggingface.co/papers/2208.12242)을 참조하세요).  사전 보존을 위해 동일한 클래스의 다른 이미지를 학습 프로세스의 일부로 사용합니다. 좋은 점은 Stable Diffusion 모델 자체를 사용하여 이러한 이미지를 생성할 수 있다는 것입니다! 학습 스크립트는 생성된 이미지를 우리가 지정한 로컬 경로에 저장합니다.

저자들에 따르면 사전 보존을 위해 `num_epochs * num_samples`개의 이미지를 생성하는 것이 좋습니다. 200-300개에서 대부분 잘 작동합니다.

<frameworkcontent>
<pt>
```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path_to_training_images"
export CLASS_DIR="path_to_class_images"
export OUTPUT_DIR="path_to_saved_model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800
```
</pt>
<jax>
```bash
export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
export INSTANCE_DIR="path-to-instance-images"
export CLASS_DIR="path-to-class-images"
export OUTPUT_DIR="path-to-save-model"

python train_dreambooth_flax.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --learning_rate=5e-6 \
  --num_class_images=200 \
  --max_train_steps=800
```
</jax>
</frameworkcontent>

## 텍스트 인코더와 and UNet로 파인튜닝하기

해당 스크립트를 사용하면 `unet`과 함께 `text_encoder`를 파인튜닝할 수 있습니다. 실험에서(자세한 내용은 [🧨 Diffusers를 사용해 DreamBooth로 Stable Diffusion 학습하기](https://huggingface.co/blog/dreambooth) 게시물을 확인하세요), 특히 얼굴 이미지를 생성할 때 훨씬 더 나은 결과를 얻을 수 있습니다.

> [!WARNING]
> 텍스트 인코더를 학습시키려면 추가 메모리가 필요해 16GB GPU로는 동작하지 않습니다. 이 옵션을 사용하려면 최소 24GB VRAM이 필요합니다.

`--train_text_encoder` 인수를 학습 스크립트에 전달하여 `text_encoder` 및 `unet`을 파인튜닝할 수 있습니다:

<frameworkcontent>
<pt>
```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path_to_training_images"
export CLASS_DIR="path_to_class_images"
export OUTPUT_DIR="path_to_saved_model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --train_text_encoder \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --use_8bit_adam
  --gradient_checkpointing \
  --learning_rate=2e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800
```
</pt>
<jax>
```bash
export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
export INSTANCE_DIR="path-to-instance-images"
export CLASS_DIR="path-to-class-images"
export OUTPUT_DIR="path-to-save-model"

python train_dreambooth_flax.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --train_text_encoder \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --learning_rate=2e-6 \
  --num_class_images=200 \
  --max_train_steps=800
```
</jax>
</frameworkcontent>

## LoRA로 파인튜닝하기

DreamBooth에서 대규모 모델의 학습을 가속화하기 위한 파인튜닝 기술인 LoRA(Low-Rank Adaptation of Large Language Models)를 사용할 수 있습니다. 자세한 내용은 [LoRA 학습](training/lora#dreambooth) 가이드를 참조하세요.

### 학습 중 체크포인트 저장하기

Dreambooth로 훈련하는 동안 과적합하기 쉬우므로, 때때로 학습 중에 정기적인 체크포인트를 저장하는 것이 유용합니다. 중간 체크포인트 중 하나가 최종 모델보다 더 잘 작동할 수 있습니다! 체크포인트 저장 기능을 활성화하려면 학습 스크립트에 다음 인수를 전달해야 합니다:

```bash
  --checkpointing_steps=500
```

이렇게 하면 `output_dir`의 하위 폴더에 전체 학습 상태가 저장됩니다. 하위 폴더 이름은 접두사 `checkpoint-`로 시작하고 지금까지 수행된 step 수입니다. 예시로 `checkpoint-1500`은 1500 학습 step 후에 저장된 체크포인트입니다.

#### 저장된 체크포인트에서 훈련 재개하기

저장된 체크포인트에서 훈련을 재개하려면, `--resume_from_checkpoint` 인수를 전달한 다음 사용할 체크포인트의 이름을 지정하면 됩니다. 특수 문자열 `"latest"`를 사용하여 저장된 마지막 체크포인트(즉, step 수가 가장 많은 체크포인트)에서 재개할 수도 있습니다. 예를 들어 다음은 1500 step 후에 저장된 체크포인트에서부터 학습을 재개합니다:

```bash
  --resume_from_checkpoint="checkpoint-1500"
```

원하는 경우 일부 하이퍼파라미터를 조정할 수 있습니다.

#### 저장된 체크포인트를 사용하여 추론 수행하기

저장된 체크포인트는 훈련 재개에 적합한 형식으로 저장됩니다. 여기에는 모델 가중치뿐만 아니라 옵티마이저, 데이터 로더 및 학습률의 상태도 포함됩니다.

**`"accelerate>=0.16.0"`**이 설치된 경우 다음 코드를 사용하여 중간 체크포인트에서 추론을 실행합니다.

```python
from diffusers import DiffusionPipeline, UNet2DConditionModel
from transformers import CLIPTextModel
import torch

# 학습에 사용된 것과 동일한 인수(model, revision)로 파이프라인을 불러옵니다.
model_id = "CompVis/stable-diffusion-v1-4"

unet = UNet2DConditionModel.from_pretrained("/sddata/dreambooth/daruma-v2-1/checkpoint-100/unet")

# `args.train_text_encoder`로 학습한 경우면 텍스트 인코더를 꼭 불러오세요
text_encoder = CLIPTextModel.from_pretrained("/sddata/dreambooth/daruma-v2-1/checkpoint-100/text_encoder")

pipeline = DiffusionPipeline.from_pretrained(model_id, unet=unet, text_encoder=text_encoder, dtype=torch.float16)
pipeline.to("cuda")

# 추론을 수행하거나 저장하거나, 허브에 푸시합니다.
pipeline.save_pretrained("dreambooth-pipeline")
```

If you have **`"accelerate<0.16.0"`** installed, you need to convert it to an inference pipeline first:

```python
from accelerate import Accelerator
from diffusers import DiffusionPipeline

# 학습에 사용된 것과 동일한 인수(model, revision)로 파이프라인을 불러옵니다.
model_id = "CompVis/stable-diffusion-v1-4"
pipeline = DiffusionPipeline.from_pretrained(model_id)

accelerator = Accelerator()

# 초기 학습에 `--train_text_encoder`가 사용된 경우 text_encoder를 사용합니다.
unet, text_encoder = accelerator.prepare(pipeline.unet, pipeline.text_encoder)

# 체크포인트 경로로부터 상태를 복원합니다. 여기서는 절대 경로를 사용해야 합니다.
accelerator.load_state("/sddata/dreambooth/daruma-v2-1/checkpoint-100")

# unwrapped 모델로 파이프라인을 다시 빌드합니다.(.unet and .text_encoder로의 할당도 작동해야 합니다)
pipeline = DiffusionPipeline.from_pretrained(
    model_id,
    unet=accelerator.unwrap_model(unet),
    text_encoder=accelerator.unwrap_model(text_encoder),
)

# 추론을 수행하거나 저장하거나, 허브에 푸시합니다.
pipeline.save_pretrained("dreambooth-pipeline")
```

## 각 GPU 용량에서의 최적화

하드웨어에 따라 16GB에서 8GB까지 GPU에서 DreamBooth를 최적화하는 몇 가지 방법이 있습니다!

### xFormers

[xFormers](https://github.com/facebookresearch/xformers)는 Transformers를 최적화하기 위한 toolbox이며, 🧨 Diffusers에서 사용되는[memory-efficient attention](https://facebookresearch.github.io/xformers/components/ops.html#module-xformers.ops)  메커니즘을 포함하고 있습니다. [xFormers를 설치](./optimization/xformers)한 다음 학습 스크립트에 다음 인수를 추가합니다:

```bash
  --enable_xformers_memory_efficient_attention
```

xFormers는 Flax에서 사용할 수 없습니다.

### 그래디언트 없음으로 설정

메모리 사용량을 줄일 수 있는 또 다른 방법은 [기울기 설정](https://pytorch.org/docs/stable/generated/torch.optim.Optimizer.zero_grad.html)을 0 대신 `None`으로 하는 것입니다. 그러나 이로 인해 특정 동작이 변경될 수 있으므로 문제가 발생하면 이 인수를 제거해 보십시오. 학습 스크립트에 다음 인수를 추가하여 그래디언트를 `None`으로 설정합니다.

```bash
  --set_grads_to_none
```

### 16GB GPU

Gradient checkpointing과 [bitsandbytes](https://github.com/TimDettmers/bitsandbytes)의 8비트 옵티마이저의 도움으로, 16GB GPU에서 dreambooth를 훈련할 수 있습니다. bitsandbytes가 설치되어 있는지 확인하세요:

```bash
pip install bitsandbytes
```

그 다음, 학습 스크립트에 `--use_8bit_adam` 옵션을 명시합니다:

```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path_to_training_images"
export CLASS_DIR="path_to_class_images"
export OUTPUT_DIR="path_to_saved_model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=2 --gradient_checkpointing \
  --use_8bit_adam \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800
```

### 12GB GPU

12GB GPU에서 DreamBooth를 실행하려면 gradient checkpointing, 8비트 옵티마이저, xFormers를 활성화하고 그래디언트를 `None`으로 설정해야 합니다.

```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path-to-instance-images"
export CLASS_DIR="path-to-class-images"
export OUTPUT_DIR="path-to-save-model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 --gradient_checkpointing \
  --use_8bit_adam \
  --enable_xformers_memory_efficient_attention \
  --set_grads_to_none \
  --learning_rate=2e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800
```

### 8GB GPU에서 학습하기

8GB GPU에 대해서는 [DeepSpeed](https://www.deepspeed.ai/)를 사용해 일부 텐서를 VRAM에서 CPU 또는 NVME로 오프로드하여 더 적은 GPU 메모리로 학습할 수도 있습니다.

🤗 Accelerate 환경을 구성하려면 다음 명령을 실행하세요:

```bash
accelerate config
```

환경 구성 중에 DeepSpeed를 사용할 것을 확인하세요.
그러면 DeepSpeed stage 2, fp16 혼합 정밀도를 결합하고 모델 매개변수와 옵티마이저 상태를 모두 CPU로 오프로드하면 8GB VRAM 미만에서 학습할 수 있습니다.
단점은 더 많은 시스템 RAM(약 25GB)이 필요하다는 것입니다. 추가 구성 옵션은 [DeepSpeed 문서](https://huggingface.co/docs/accelerate/usage_guides/deepspeed)를 참조하세요.

또한 기본 Adam 옵티마이저를 DeepSpeed의 최적화된 Adam 버전으로 변경해야 합니다.
이는 상당한 속도 향상을 위한 Adam인 [`deepspeed.ops.adam.DeepSpeedCPUAdam`](https://deepspeed.readthedocs.io/en/latest/optimizers.html#adam-cpu)입니다.
`DeepSpeedCPUAdam`을 활성화하려면 시스템의 CUDA toolchain 버전이 PyTorch와 함께 설치된 것과 동일해야 합니다.

8비트 옵티마이저는 현재 DeepSpeed와 호환되지 않는 것 같습니다.

다음 명령으로 학습을 시작합니다:

```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export INSTANCE_DIR="path_to_training_images"
export CLASS_DIR="path_to_class_images"
export OUTPUT_DIR="path_to_saved_model"

accelerate launch train_dreambooth.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --instance_data_dir=$INSTANCE_DIR \
  --class_data_dir=$CLASS_DIR \
  --output_dir=$OUTPUT_DIR \
  --with_prior_preservation --prior_loss_weight=1.0 \
  --instance_prompt="a photo of sks dog" \
  --class_prompt="a photo of dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --sample_batch_size=1 \
  --gradient_accumulation_steps=1 --gradient_checkpointing \
  --learning_rate=5e-6 \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --num_class_images=200 \
  --max_train_steps=800 \
  --mixed_precision=fp16
```

## 추론

모델을 학습한 후에는, 모델이 저장된 경로를 지정해 `StableDiffusionPipeline`로 추론을 수행할 수 있습니다. 프롬프트에 학습에 사용된 특수 `식별자`(이전 예시의 `sks`)가 포함되어 있는지 확인하세요.

**`"accelerate>=0.16.0"`**이 설치되어 있는 경우 다음 코드를 사용하여 중간 체크포인트에서 추론을 실행할 수 있습니다:

```python
from diffusers import StableDiffusionPipeline
import torch

model_id = "path_to_saved_model"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")

prompt = "A photo of sks dog in a bucket"
image = pipe(prompt, num_inference_steps=50, guidance_scale=7.5).images[0]

image.save("dog-bucket.png")
```

[저장된 학습 체크포인트](#inference-from-a-saved-checkpoint)에서도 추론을 실행할 수도 있습니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/dreambooth.md" />

### Low-Rank Adaptation of Large Language Models (LoRA)
https://huggingface.co/docs/diffusers/main/ko/training/lora.md

# Low-Rank Adaptation of Large Language Models (LoRA)


> [!WARNING]
> 현재 LoRA는 `UNet2DConditionalModel`의 어텐션 레이어에서만 지원됩니다.

[LoRA(Low-Rank Adaptation of Large Language Models)](https://huggingface.co/papers/2106.09685)는 메모리를 적게 사용하면서 대규모 모델의 학습을 가속화하는 학습 방법입니다. 이는 rank-decomposition weight 행렬 쌍(**업데이트 행렬**이라고 함)을 추가하고 새로 추가된 가중치**만** 학습합니다. 여기에는 몇 가지 장점이 있습니다.

- 이전에 미리 학습된 가중치는 고정된 상태로 유지되므로 모델이 [치명적인 망각](https://www.pnas.org/doi/10.1073/pnas.1611835114) 경향이 없습니다.
- Rank-decomposition 행렬은 원래 모델보다 파라메터 수가 훨씬 적으므로 학습된 LoRA 가중치를 쉽게 끼워넣을 수 있습니다.
- LoRA 매트릭스는 일반적으로 원본 모델의 어텐션 레이어에 추가됩니다. 🧨 Diffusers는 `load_attn_procs()` 메서드를 제공하여 LoRA 가중치를 모델의 어텐션 레이어로 불러옵니다. `scale` 매개변수를 통해 모델이 새로운 학습 이미지에 맞게 조정되는 범위를 제어할 수 있습니다.
- 메모리 효율성이 향상되어 Tesla T4, RTX 3080 또는 RTX 2080 Ti와 같은 소비자용 GPU에서 파인튜닝을 실행할 수 있습니다! T4와 같은 GPU는 무료이며 Kaggle 또는 Google Colab 노트북에서 쉽게 액세스할 수 있습니다.


> [!TIP]
> 💡 LoRA는 어텐션 레이어에만 한정되지는 않습니다. 저자는 언어 모델의 어텐션 레이어를 수정하는 것이 매우 효율적으로 죻은 성능을 얻기에 충분하다는 것을 발견했습니다. 이것이 LoRA 가중치를 모델의 어텐션 레이어에 추가하는 것이 일반적인 이유입니다. LoRA 작동 방식에 대한 자세한 내용은 [Using LoRA for effective Stable Diffusion fine-tuning](https://huggingface.co/blog/lora) 블로그를 확인하세요!

[cloneofsimo](https://github.com/cloneofsimo)는 인기 있는 [lora](https://github.com/cloneofsimo/lora) GitHub 리포지토리에서 Stable Diffusion을 위한 LoRA 학습을 최초로 시도했습니다. 🧨 Diffusers는 [text-to-image 생성](https://github.com/huggingface/diffusers/tree/main/examples/text_to_image#training-with-lora) 및 [DreamBooth](https://github.com/huggingface/diffusers/tree/main/examples/dreambooth#training-with-low-rank-adaptation-of-large-language-models-lora)을 지원합니다. 이 가이드는 두 가지를 모두 수행하는 방법을 보여줍니다.

모델을 저장하거나 커뮤니티와 공유하려면 Hugging Face 계정에 로그인하세요(아직 계정이 없는 경우 [생성](https://huggingface.co/join)하세요):

```bash
hf auth login
```

## Text-to-image

수십억 개의 파라메터들이 있는 Stable Diffusion과 같은 모델을 파인튜닝하는 것은 느리고 어려울 수 있습니다. LoRA를 사용하면 diffusion 모델을 파인튜닝하는 것이 훨씬 쉽고 빠릅니다. 8비트 옵티마이저와 같은 트릭에 의존하지 않고도 11GB의 GPU RAM으로 하드웨어에서 실행할 수 있습니다.


### 학습[[dreambooth-training]]

[Naruto BLIP 캡션](https://huggingface.co/datasets/lambdalabs/naruto-blip-captions) 데이터셋으로 [`stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)를 파인튜닝해 나만의 포켓몬을 생성해 보겠습니다.

시작하려면 `MODEL_NAME` 및 `DATASET_NAME` 환경 변수가 설정되어 있는지 확인하십시오. `OUTPUT_DIR` 및 `HUB_MODEL_ID` 변수는 선택 사항이며 허브에서 모델을 저장할 위치를 지정합니다.

```bash
export MODEL_NAME="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="/sddata/finetune/lora/naruto"
export HUB_MODEL_ID="naruto-lora"
export DATASET_NAME="lambdalabs/naruto-blip-captions"
```

학습을 시작하기 전에 알아야 할 몇 가지 플래그가 있습니다.

* `--push_to_hub`를 명시하면 학습된 LoRA 임베딩을 허브에 저장합니다.
* `--report_to=wandb`는 학습 결과를 가중치 및 편향 대시보드에 보고하고 기록합니다(예를 들어, 이 [보고서](https://wandb.ai/pcuenq/text2image-fine-tune/run/b4k1w0tn?workspace=user-pcuenq)를 참조하세요).
* `--learning_rate=1e-04`, 일반적으로 LoRA에서 사용하는 것보다 더 높은 학습률을 사용할 수 있습니다.

이제 학습을 시작할 준비가 되었습니다 (전체 학습 스크립트는 [여기](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_lora.py)에서 찾을 수 있습니다).

```bash
accelerate launch train_dreambooth_lora.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --instance_prompt="a photo of sks dog" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=1 \
  --checkpointing_steps=100 \
  --learning_rate=1e-4 \
  --report_to="wandb" \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --max_train_steps=500 \
  --validation_prompt="A photo of sks dog in a bucket" \
  --validation_epochs=50 \
  --seed="0" \
  --push_to_hub
```

### 추론[[dreambooth-inference]]

이제 `StableDiffusionPipeline`에서 기본 모델을 불러와 추론을 위해 모델을 사용할 수 있습니다:

```py
>>> import torch
>>> from diffusers import StableDiffusionPipeline

>>> model_base = "stable-diffusion-v1-5/stable-diffusion-v1-5"

>>> pipe = StableDiffusionPipeline.from_pretrained(model_base, torch_dtype=torch.float16)
```

*기본 모델의 가중치 위에* 파인튜닝된 DreamBooth 모델에서 LoRA 가중치를 불러온 다음, 더 빠른 추론을 위해 파이프라인을 GPU로 이동합니다. LoRA 가중치를 프리징된 사전 훈련된 모델 가중치와 병합할 때, 선택적으로 'scale' 매개변수로 어느 정도의 가중치를 병합할 지 조절할 수 있습니다:

> [!TIP]
> 💡 `0`의 `scale` 값은 LoRA 가중치를 사용하지 않아 원래 모델의 가중치만 사용한 것과 같고, `1`의 `scale` 값은 파인튜닝된 LoRA 가중치만 사용함을 의미합니다. 0과 1 사이의 값들은 두 결과들 사이로 보간됩니다.

```py
>>> pipe.unet.load_attn_procs(model_path)
>>> pipe.to("cuda")
# LoRA 파인튜닝된 모델의 가중치 절반과 기본 모델의 가중치 절반 사용

>>> image = pipe(
...     "A picture of a sks dog in a bucket.",
...     num_inference_steps=25,
...     guidance_scale=7.5,
...     cross_attention_kwargs={"scale": 0.5},
... ).images[0]
# 완전히 파인튜닝된 LoRA 모델의 가중치 사용

>>> image = pipe("A picture of a sks dog in a bucket.", num_inference_steps=25, guidance_scale=7.5).images[0]
>>> image.save("bucket-dog.png")
```

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/lora.md" />

### 여러 GPU를 사용한 분산 추론
https://huggingface.co/docs/diffusers/main/ko/training/distributed_inference.md

# 여러 GPU를 사용한 분산 추론

분산 설정에서는 여러 개의 프롬프트를 동시에 생성할 때 유용한 🤗 [Accelerate](https://huggingface.co/docs/accelerate/index) 또는 [PyTorch Distributed](https://pytorch.org/tutorials/beginner/dist_overview.html)를 사용하여 여러 GPU에서 추론을 실행할 수 있습니다.

이 가이드에서는 분산 추론을 위해 🤗 Accelerate와 PyTorch Distributed를 사용하는 방법을 보여드립니다.

## 🤗 Accelerate

🤗 [Accelerate](https://huggingface.co/docs/accelerate/index)는 분산 설정에서 추론을 쉽게 훈련하거나 실행할 수 있도록 설계된 라이브러리입니다. 분산 환경 설정 프로세스를 간소화하여 PyTorch 코드에 집중할 수 있도록 해줍니다.

시작하려면 Python 파일을 생성하고 `accelerate.PartialState`를 초기화하여 분산 환경을 생성하면, 설정이 자동으로 감지되므로 `rank` 또는 `world_size`를 명시적으로 정의할 필요가 없습니다. ['DiffusionPipeline`]을 `distributed_state.device`로 이동하여 각 프로세스에 GPU를 할당합니다.

이제 컨텍스트 관리자로 `split_between_processes` 유틸리티를 사용하여 프로세스 수에 따라 프롬프트를 자동으로 분배합니다.


```py
from accelerate import PartialState
from diffusers import DiffusionPipeline

pipeline = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16)
distributed_state = PartialState()
pipeline.to(distributed_state.device)

with distributed_state.split_between_processes(["a dog", "a cat"]) as prompt:
    result = pipeline(prompt).images[0]
    result.save(f"result_{distributed_state.process_index}.png")
```

Use the `--num_processes` argument to specify the number of GPUs to use, and call `accelerate launch` to run the script:

```bash
accelerate launch run_distributed.py --num_processes=2
```

> [!TIP]
> 자세한 내용은 [🤗 Accelerate를 사용한 분산 추론](https://huggingface.co/docs/accelerate/en/usage_guides/distributed_inference#distributed-inference-with-accelerate) 가이드를 참조하세요.

## Pytoerch 분산

PyTorch는 데이터 병렬 처리를 가능하게 하는 [`DistributedDataParallel`](https://pytorch.org/docs/stable/generated/torch.nn.parallel.DistributedDataParallel.html)을 지원합니다.

시작하려면 Python 파일을 생성하고 `torch.distributed` 및 `torch.multiprocessing`을 임포트하여 분산 프로세스 그룹을 설정하고 각 GPU에서 추론용 프로세스를 생성합니다. 그리고 `DiffusionPipeline`도 초기화해야 합니다:

확산 파이프라인을 `rank`로 이동하고 `get_rank`를 사용하여 각 프로세스에 GPU를 할당하면 각 프로세스가 다른 프롬프트를 처리합니다:

```py
import torch
import torch.distributed as dist
import torch.multiprocessing as mp

from diffusers import DiffusionPipeline

sd = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16)
```

사용할 백엔드 유형, 현재 프로세스의 `rank`, `world_size` 또는 참여하는 프로세스 수로 분산 환경 생성을 처리하는 함수`init_process_group`를 만들어 추론을 실행해야 합니다.

2개의 GPU에서 추론을 병렬로 실행하는 경우 `world_size`는 2입니다.

```py
def run_inference(rank, world_size):
    dist.init_process_group("nccl", rank=rank, world_size=world_size)

    sd.to(rank)

    if torch.distributed.get_rank() == 0:
        prompt = "a dog"
    elif torch.distributed.get_rank() == 1:
        prompt = "a cat"

    image = sd(prompt).images[0]
    image.save(f"./{'_'.join(prompt)}.png")
```

분산 추론을 실행하려면 [`mp.spawn`](https://pytorch.org/docs/stable/multiprocessing.html#torch.multiprocessing.spawn)을 호출하여 `world_size`에 정의된 GPU 수에 대해 `run_inference` 함수를 실행합니다:

```py
def main():
    world_size = 2
    mp.spawn(run_inference, args=(world_size,), nprocs=world_size, join=True)


if __name__ == "__main__":
    main()
```

추론 스크립트를 완료했으면 `--nproc_per_node` 인수를 사용하여 사용할 GPU 수를 지정하고 `torchrun`을 호출하여 스크립트를 실행합니다:

```bash
torchrun run_distributed.py --nproc_per_node=2
```

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/distributed_inference.md" />

### 🧨 Diffusers 학습 예시
https://huggingface.co/docs/diffusers/main/ko/training/overview.md

# 🧨 Diffusers 학습 예시

이번 챕터에서는 다양한 유즈케이스들에 대한 예제 코드들을 통해 어떻게하면 효과적으로 `diffusers` 라이브러리를 사용할 수 있을까에 대해 알아보도록 하겠습니다.

**Note**: 혹시 오피셜한 예시코드를 찾고 있다면, [여기](https://github.com/huggingface/diffusers/tree/main/src/diffusers/pipelines)를 참고해보세요!

여기서 다룰 예시들은 다음을 지향합니다.

- **손쉬운 디펜던시 설치** (Self-contained) : 여기서 사용될 예시 코드들의 디펜던시 패키지들은 전부 `pip install` 명령어를 통해 설치 가능한 패키지들입니다. 또한 친절하게 `requirements.txt` 파일에 해당 패키지들이 명시되어 있어, `pip install -r requirements.txt`로 간편하게 해당 디펜던시들을 설치할 수 있습니다. 예시: [train_unconditional.py](https://github.com/huggingface/diffusers/blob/main/examples/unconditional_image_generation/train_unconditional.py), [requirements.txt](https://github.com/huggingface/diffusers/blob/main/examples/unconditional_image_generation/requirements.txt)
- **손쉬운 수정** (Easy-to-tweak) : 저희는 가능하면 많은 유즈 케이스들을 제공하고자 합니다. 하지만 예시는 결국 그저 예시라는 점들 기억해주세요. 여기서 제공되는 예시코드들을 그저 단순히 복사-붙혀넣기하는 식으로는 여러분이 마주한 문제들을 손쉽게 해결할 순 없을 것입니다. 다시 말해 어느 정도는 여러분의 상황과 니즈에 맞춰 코드를 일정 부분 고쳐나가야 할 것입니다. 따라서 대부분의 학습 예시들은 데이터의 전처리 과정과 학습 과정에 대한 코드들을 함께 제공함으로써, 사용자가 니즈에 맞게 손쉬운 수정할 수 있도록 돕고 있습니다.
- **입문자 친화적인** (Beginner-friendly) : 이번 챕터는 diffusion 모델과 `diffusers` 라이브러리에 대한 전반적인 이해를 돕기 위해 작성되었습니다. 따라서 diffusion 모델에 대한 최신 SOTA (state-of-the-art) 방법론들 가운데서도, 입문자에게는 많이 어려울 수 있다고 판단되면, 해당 방법론들은 여기서 다루지 않으려고 합니다.
- **하나의 태스크만 포함할 것**(One-purpose-only): 여기서 다룰 예시들은 하나의 태스크만 포함하고 있어야 합니다. 물론 이미지 초해상화(super-resolution)와 이미지 보정(modification)과 같은 유사한 모델링 프로세스를 갖는 태스크들이 존재하겠지만, 하나의 예제에 하나의 태스크만을 담는 것이 더 이해하기 용이하다고 판단했기 때문입니다.



저희는 diffusion 모델의 대표적인 태스크들을 다루는 공식 예제를 제공하고 있습니다. *공식* 예제는 현재 진행형으로 `diffusers` 관리자들(maintainers)에 의해 관리되고 있습니다. 또한 저희는 앞서 정의한 저희의 철학을 엄격하게 따르고자 노력하고 있습니다. 혹시 여러분께서 이러한 예시가 반드시 필요하다고 생각되신다면, 언제든지 [Feature Request](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feature_request.md&title=) 혹은 직접 [Pull Request](https://github.com/huggingface/diffusers/compare)를 주시기 바랍니다. 저희는 언제나 환영입니다!

학습 예시들은 다양한 태스크들에 대해 diffusion 모델을 사전학습(pretrain)하거나 파인튜닝(fine-tuning)하는 법을 보여줍니다. 현재 다음과 같은 예제들을 지원하고 있습니다.

- [Unconditional Training](./unconditional_training)
- [Text-to-Image Training](./text2image)
- [Text Inversion](./text_inversion)
- [Dreambooth](./dreambooth)

memory-efficient attention 연산을 수행하기 위해, 가능하면 [xFormers](../optimization/xformers)를 설치해주시기 바랍니다. 이를 통해 학습 속도를 늘리고 메모리에 대한 부담을 줄일 수 있습니다.

| Task | 🤗 Accelerate | 🤗 Datasets | Colab
|---|---|:---:|:---:|
| [**Unconditional Image Generation**](./unconditional_training) | ✅ | ✅ | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb)
| [**Text-to-Image fine-tuning**](./text2image) | ✅ | ✅ |
| [**Textual Inversion**](./text_inversion) | ✅ | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_textual_inversion_training.ipynb)
| [**Dreambooth**](./dreambooth) | ✅ | - | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/sd_dreambooth_training.ipynb)
| [**Training with LoRA**](./lora) | ✅ | - | - |
| [**ControlNet**](./controlnet) | ✅ | ✅ | - |
| [**InstructPix2Pix**](./instructpix2pix) | ✅ | ✅ | - |
| [**Custom Diffusion**](./custom_diffusion) | ✅ | ✅ | - |


## 커뮤니티

공식 예제 외에도 **커뮤니티 예제** 역시 제공하고 있습니다. 해당 예제들은 우리의 커뮤니티에 의해 관리됩니다. 커뮤니티 예쩨는 학습 예시나 추론 파이프라인으로 구성될 수 있습니다. 이러한 커뮤니티 예시들의 경우,  앞서 정의했던 철학들을 좀 더 관대하게 적용하고 있습니다. 또한 이러한 커뮤니티 예시들의 경우, 모든 이슈들에 대한 유지보수를 보장할 수는 없습니다.

유용하긴 하지만, 아직은 대중적이지 못하거나 저희의 철학에 부합하지 않는 예제들은 [community examples](https://github.com/huggingface/diffusers/tree/main/examples/community) 폴더에 담기게 됩니다.

**Note**: 커뮤니티 예제는 `diffusers`에 기여(contribution)를 희망하는 분들에게 [아주 좋은 기여 수단](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22)이 될 수 있습니다.

## 주목할 사항들

최신 버전의 예시 코드들의 성공적인 구동을 보장하기 위해서는, 반드시 **소스코드를 통해 `diffusers`를 설치해야 하며,** 해당 예시 코드들이 요구하는 디펜던시들 역시 설치해야 합니다. 이를 위해 새로운 가상 환경을 구축하고 다음의 명령어를 실행해야 합니다.

```bash
git clone https://github.com/huggingface/diffusers
cd diffusers
pip install .
```

그 다음 `cd` 명령어를 통해 해당 예제 디렉토리에 접근해서 다음 명령어를 실행하면 됩니다.

```bash
pip install -r requirements.txt
```

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/overview.md" />

### Text-to-image
https://huggingface.co/docs/diffusers/main/ko/training/text2image.md

# Text-to-image

> [!WARNING]
> text-to-image 파인튜닝 스크립트는 experimental 상태입니다. 과적합하기 쉽고 치명적인 망각과 같은 문제에 부딪히기 쉽습니다. 자체 데이터셋에서 최상의 결과를 얻으려면 다양한 하이퍼파라미터를 탐색하는 것이 좋습니다.

Stable Diffusion과 같은 text-to-image 모델은 텍스트 프롬프트에서 이미지를 생성합니다. 이 가이드는 PyTorch 및 Flax를 사용하여 자체 데이터셋에서 [`CompVis/stable-diffusion-v1-4`](https://huggingface.co/CompVis/stable-diffusion-v1-4) 모델로 파인튜닝하는 방법을 보여줍니다. 이 가이드에 사용된 text-to-image 파인튜닝을 위한 모든 학습 스크립트에 관심이 있는 경우 이 [리포지토리](https://github.com/huggingface/diffusers/tree/main/examples/text_to_image)에서 자세히 찾을 수 있습니다.

스크립트를 실행하기 전에, 라이브러리의 학습 dependency들을 설치해야 합니다:

```bash
pip install git+https://github.com/huggingface/diffusers.git
pip install -U -r requirements.txt
```

그리고 [🤗Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화합니다:

```bash
accelerate config
```

리포지토리를 이미 복제한 경우, 이 단계를 수행할 필요가 없습니다. 대신, 로컬 체크아웃 경로를 학습 스크립트에 명시할 수 있으며 거기에서 로드됩니다.

### 하드웨어 요구 사항

`gradient_checkpointing` 및 `mixed_precision`을 사용하면 단일 24GB GPU에서 모델을 파인튜닝할 수 있습니다. 더 높은 `batch_size`와 더 빠른 훈련을 위해서는 GPU 메모리가 30GB 이상인 GPU를 사용하는 것이 좋습니다. TPU 또는 GPU에서 파인튜닝을 위해 JAX나 Flax를 사용할 수도 있습니다. 자세한 내용은 [아래](#flax-jax-finetuning)를 참조하세요.

xFormers로 memory efficient attention을 활성화하여 메모리 사용량 훨씬 더 줄일 수 있습니다. [xFormers가 설치](./optimization/xformers)되어 있는지 확인하고 `--enable_xformers_memory_efficient_attention`를 학습 스크립트에 명시합니다.

xFormers는 Flax에 사용할 수 없습니다.

## Hub에 모델 업로드하기

학습 스크립트에 다음 인수를 추가하여 모델을 허브에 저장합니다:

```bash
  --push_to_hub
```


## 체크포인트 저장 및 불러오기

학습 중 발생할 수 있는 일에 대비하여 정기적으로 체크포인트를 저장해 두는 것이 좋습니다. 체크포인트를 저장하려면 학습 스크립트에 다음 인수를 명시합니다.

```bash
  --checkpointing_steps=500
```

500스텝마다 전체 학습 state가 'output_dir'의 하위 폴더에 저장됩니다. 체크포인트는 'checkpoint-'에 지금까지 학습된 step 수입니다. 예를 들어 'checkpoint-1500'은 1500 학습 step 후에 저장된 체크포인트입니다.

학습을 재개하기 위해 체크포인트를 불러오려면 '--resume_from_checkpoint' 인수를 학습 스크립트에 명시하고 재개할 체크포인트를 지정하십시오. 예를 들어 다음 인수는 1500개의 학습 step 후에 저장된 체크포인트에서부터 훈련을 재개합니다.

```bash
  --resume_from_checkpoint="checkpoint-1500"
```

## 파인튜닝

<frameworkcontent>
<pt>
다음과 같이 [Naruto BLIP 캡션](https://huggingface.co/datasets/lambdalabs/naruto-blip-captions) 데이터셋에서 파인튜닝 실행을 위해 [PyTorch 학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image.py)를 실행합니다:


```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export dataset_name="lambdalabs/naruto-blip-captions"

accelerate launch train_text_to_image.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --dataset_name=$dataset_name \
  --use_ema \
  --resolution=512 --center_crop --random_flip \
  --train_batch_size=1 \
  --gradient_accumulation_steps=4 \
  --gradient_checkpointing \
  --mixed_precision="fp16" \
  --max_train_steps=15000 \
  --learning_rate=1e-05 \
  --max_grad_norm=1 \
  --lr_scheduler="constant" --lr_warmup_steps=0 \
  --output_dir="sd-naruto-model"
```

자체 데이터셋으로 파인튜닝하려면 🤗 [Datasets](https://huggingface.co/docs/datasets/index)에서 요구하는 형식에 따라 데이터셋을 준비하세요. [데이터셋을 허브에 업로드](https://huggingface.co/docs/datasets/image_dataset#upload-dataset-to-the-hub)하거나 [파일들이 있는 로컬 폴더를 준비](https ://huggingface.co/docs/datasets/image_dataset#imagefolder)할 수 있습니다.

사용자 커스텀 loading logic을 사용하려면 스크립트를 수정하십시오. 도움이 되도록 코드의 적절한 위치에 포인터를 남겼습니다. 🤗 아래 예제 스크립트는 `TRAIN_DIR`의 로컬 데이터셋으로를 파인튜닝하는 방법과 `OUTPUT_DIR`에서 모델을 저장할 위치를 보여줍니다:


```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export TRAIN_DIR="path_to_your_dataset"
export OUTPUT_DIR="path_to_save_model"

accelerate launch train_text_to_image.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$TRAIN_DIR \
  --use_ema \
  --resolution=512 --center_crop --random_flip \
  --train_batch_size=1 \
  --gradient_accumulation_steps=4 \
  --gradient_checkpointing \
  --mixed_precision="fp16" \
  --max_train_steps=15000 \
  --learning_rate=1e-05 \
  --max_grad_norm=1 \
  --lr_scheduler="constant" --lr_warmup_steps=0 \
  --output_dir=${OUTPUT_DIR}
```

</pt>
<jax>
[@duongna211](https://github.com/duongna21)의 기여로, Flax를 사용해 TPU 및 GPU에서 Stable Diffusion 모델을 더 빠르게 학습할 수 있습니다. 이는 TPU 하드웨어에서 매우 효율적이지만 GPU에서도 훌륭하게 작동합니다. Flax 학습 스크립트는 gradient checkpointing나 gradient accumulation과 같은 기능을 아직 지원하지 않으므로 메모리가 30GB 이상인 GPU 또는 TPU v3가 필요합니다.

스크립트를 실행하기 전에 요구 사항이 설치되어 있는지 확인하십시오:

```bash
pip install -U -r requirements_flax.txt
```

그러면 다음과 같이 [Flax 학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/text_to_image/train_text_to_image_flax.py)를 실행할 수 있습니다.

```bash
export MODEL_NAME="stable-diffusion-v1-5/stable-diffusion-v1-5"
export dataset_name="lambdalabs/naruto-blip-captions"

python train_text_to_image_flax.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --dataset_name=$dataset_name \
  --resolution=512 --center_crop --random_flip \
  --train_batch_size=1 \
  --max_train_steps=15000 \
  --learning_rate=1e-05 \
  --max_grad_norm=1 \
  --output_dir="sd-naruto-model"
```

자체 데이터셋으로 파인튜닝하려면 🤗 [Datasets](https://huggingface.co/docs/datasets/index)에서 요구하는 형식에 따라 데이터셋을 준비하세요. [데이터셋을 허브에 업로드](https://huggingface.co/docs/datasets/image_dataset#upload-dataset-to-the-hub)하거나 [파일들이 있는 로컬 폴더를 준비](https ://huggingface.co/docs/datasets/image_dataset#imagefolder)할 수 있습니다.

사용자 커스텀 loading logic을 사용하려면 스크립트를 수정하십시오. 도움이 되도록 코드의 적절한 위치에 포인터를 남겼습니다. 🤗 아래 예제 스크립트는 `TRAIN_DIR`의 로컬 데이터셋으로를 파인튜닝하는 방법을 보여줍니다:

```bash
export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
export TRAIN_DIR="path_to_your_dataset"

python train_text_to_image_flax.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$TRAIN_DIR \
  --resolution=512 --center_crop --random_flip \
  --train_batch_size=1 \
  --mixed_precision="fp16" \
  --max_train_steps=15000 \
  --learning_rate=1e-05 \
  --max_grad_norm=1 \
  --output_dir="sd-naruto-model"
```
</jax>
</frameworkcontent>

## LoRA

Text-to-image 모델 파인튜닝을 위해, 대규모 모델 학습을 가속화하기 위한 파인튜닝 기술인 LoRA(Low-Rank Adaptation of Large Language Models)를 사용할 수 있습니다. 자세한 내용은 [LoRA 학습](lora#text-to-image) 가이드를 참조하세요.

## 추론

허브의 모델 경로 또는 모델 이름을 `StableDiffusionPipeline`에 전달하여 추론을 위해 파인 튜닝된 모델을 불러올 수 있습니다:

<frameworkcontent>
<pt>
```python
from diffusers import StableDiffusionPipeline

model_path = "path_to_saved_model"
pipe = StableDiffusionPipeline.from_pretrained(model_path, torch_dtype=torch.float16)
pipe.to("cuda")

image = pipe(prompt="yoda").images[0]
image.save("yoda-naruto.png")
```
</pt>
<jax>
```python
import jax
import numpy as np
from flax.jax_utils import replicate
from flax.training.common_utils import shard
from diffusers import FlaxStableDiffusionPipeline

model_path = "path_to_saved_model"
pipe, params = FlaxStableDiffusionPipeline.from_pretrained(model_path, dtype=jax.numpy.bfloat16)

prompt = "yoda naruto"
prng_seed = jax.random.PRNGKey(0)
num_inference_steps = 50

num_samples = jax.device_count()
prompt = num_samples * [prompt]
prompt_ids = pipeline.prepare_inputs(prompt)

# shard inputs and rng
params = replicate(params)
prng_seed = jax.random.split(prng_seed, jax.device_count())
prompt_ids = shard(prompt_ids)

images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
image.save("yoda-naruto.png")
```
</jax>
</frameworkcontent>

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/text2image.md" />

### Unconditional 이미지 생성
https://huggingface.co/docs/diffusers/main/ko/training/unconditional_training.md

# Unconditional 이미지 생성

unconditional 이미지 생성은 text-to-image 또는 image-to-image 모델과 달리 텍스트나 이미지에 대한 조건이 없이 학습 데이터 분포와 유사한 이미지만을 생성합니다.

<iframe
	src="https://stevhliu-ddpm-butterflies-128.hf.space"
	frameborder="0"
	width="850"
	height="550"
></iframe>


이 가이드에서는 기존에 존재하던 데이터셋과 자신만의 커스텀 데이터셋에 대해 unconditional image generation 모델을 훈련하는 방법을 설명합니다. 훈련 세부 사항에 대해 더 자세히 알고 싶다면 unconditional image generation을 위한 모든 학습 스크립트를 [여기](https://github.com/huggingface/diffusers/tree/main/examples/unconditional_image_generation)에서 확인할 수 있습니다.

스크립트를 실행하기 전, 먼저 의존성 라이브러리들을 설치해야 합니다.

```bash
pip install diffusers[training] accelerate datasets
```

그 다음 🤗 [Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화합니다.

```bash
accelerate config
```

별도의 설정 없이 기본 설정으로 🤗 [Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화해봅시다.

```bash
accelerate config default
```

노트북과 같은 대화형 쉘을 지원하지 않는 환경의 경우, 다음과 같이 사용해볼 수도 있습니다.

```py
from accelerate.utils import write_basic_config

write_basic_config()
```

## 모델을 허브에 업로드하기

학습 스크립트에 다음 인자를 추가하여 허브에 모델을 업로드할 수 있습니다.

```bash
--push_to_hub
```

## 체크포인트 저장하고 불러오기

훈련 중 문제가 발생할 경우를 대비하여 체크포인트를 정기적으로 저장하는 것이 좋습니다. 체크포인트를 저장하려면 학습 스크립트에 다음 인자를 전달합니다:

```bash
--checkpointing_steps=500
```

전체 훈련 상태는 500스텝마다 `output_dir`의 하위 폴더에 저장되며, 학습 스크립트에 `--resume_from_checkpoint` 인자를 전달함으로써 체크포인트를 불러오고 훈련을 재개할 수 있습니다.

```bash
--resume_from_checkpoint="checkpoint-1500"
```

## 파인튜닝

이제 학습 스크립트를 시작할 준비가 되었습니다! `--dataset_name` 인자에 파인튜닝할 데이터셋 이름을 지정한 다음, `--output_dir` 인자에 지정된 경로로 저장합니다. 본인만의 데이터셋를 사용하려면, [학습용 데이터셋 만들기](create_dataset) 가이드를 참조하세요.

학습 스크립트는 `diffusion_pytorch_model.bin` 파일을 생성하고, 그것을 당신의 리포지토리에 저장합니다.

> [!TIP]
> 💡 전체 학습은 V100 GPU 4개를 사용할 경우, 2시간이 소요됩니다.

예를 들어, [Oxford Flowers](https://huggingface.co/datasets/huggan/flowers-102-categories) 데이터셋을 사용해 파인튜닝할 경우:

```bash
accelerate launch train_unconditional.py \
  --dataset_name="huggan/flowers-102-categories" \
  --resolution=64 \
  --output_dir="ddpm-ema-flowers-64" \
  --train_batch_size=16 \
  --num_epochs=100 \
  --gradient_accumulation_steps=1 \
  --learning_rate=1e-4 \
  --lr_warmup_steps=500 \
  --mixed_precision=no \
  --push_to_hub
```

<div class="flex justify-center">
    <img src="https://user-images.githubusercontent.com/26864830/180248660-a0b143d0-b89a-42c5-8656-2ebf6ece7e52.png"/>
</div>
[Naruto](https://huggingface.co/datasets/lambdalabs/naruto-blip-captions) 데이터셋을 사용할 경우:

```bash
accelerate launch train_unconditional.py \
  --dataset_name="lambdalabs/naruto-blip-captions" \
  --resolution=64 \
  --output_dir="ddpm-ema-naruto-64" \
  --train_batch_size=16 \
  --num_epochs=100 \
  --gradient_accumulation_steps=1 \
  --learning_rate=1e-4 \
  --lr_warmup_steps=500 \
  --mixed_precision=no \
  --push_to_hub
```

<div class="flex justify-center">
    <img src="https://user-images.githubusercontent.com/26864830/180248200-928953b4-db38-48db-b0c6-8b740fe6786f.png"/>
</div>

### 여러개의 GPU로 훈련하기

`accelerate`을 사용하면 원활한 다중 GPU 훈련이 가능합니다. `accelerate`을 사용하여 분산 훈련을 실행하려면 [여기](https://huggingface.co/docs/accelerate/basic_tutorials/launch) 지침을 따르세요. 다음은 명령어 예제입니다.

```bash
accelerate launch --mixed_precision="fp16" --multi_gpu train_unconditional.py \
  --dataset_name="lambdalabs/naruto-blip-captions" \
  --resolution=64 --center_crop --random_flip \
  --output_dir="ddpm-ema-naruto-64" \
  --train_batch_size=16 \
  --num_epochs=100 \
  --gradient_accumulation_steps=1 \
  --use_ema \
  --learning_rate=1e-4 \
  --lr_warmup_steps=500 \
  --mixed_precision="fp16" \
  --logger="wandb" \
  --push_to_hub
```


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/unconditional_training.md" />

### 커스텀 Diffusion 학습 예제
https://huggingface.co/docs/diffusers/main/ko/training/custom_diffusion.md

# 커스텀 Diffusion 학습 예제

[커스텀 Diffusion](https://huggingface.co/papers/2212.04488)은 피사체의 이미지 몇 장(4~5장)만 주어지면 Stable Diffusion처럼 text-to-image 모델을 커스터마이징하는 방법입니다.
'train_custom_diffusion.py' 스크립트는 학습 과정을 구현하고 이를 Stable Diffusion에 맞게 조정하는 방법을 보여줍니다.

이 교육 사례는 [Nupur Kumari](https://nupurkmr9.github.io/)가 제공하였습니다. (Custom Diffusion의 저자 중 한명).

## 로컬에서 PyTorch로 실행하기

### Dependencies 설치하기

스크립트를 실행하기 전에 라이브러리의 학습 dependencies를 설치해야 합니다:

**중요**

예제 스크립트의 최신 버전을 성공적으로 실행하려면 **소스로부터 설치**하는 것을 매우 권장하며, 예제 스크립트를 자주 업데이트하는 만큼 일부 예제별 요구 사항을 설치하고 설치를 최신 상태로 유지하는 것이 좋습니다. 이를 위해 새 가상 환경에서 다음 단계를 실행하세요:


```bash
git clone https://github.com/huggingface/diffusers
cd diffusers
pip install -e .
```

[example folder](https://github.com/huggingface/diffusers/tree/main/examples/custom_diffusion)로 cd하여 이동하세요.

```
cd examples/custom_diffusion
```

이제 실행

```bash
pip install -r requirements.txt
pip install clip-retrieval
```

그리고 [🤗Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화:

```bash
accelerate config
```

또는 사용자 환경에 대한 질문에 답하지 않고 기본 가속 구성을 사용하려면 다음과 같이 하세요.

```bash
accelerate config default
```

또는 사용 중인 환경이 대화형 셸을 지원하지 않는 경우(예: jupyter notebook)

```python
from accelerate.utils import write_basic_config

write_basic_config()
```
### 고양이 예제 😺

이제 데이터셋을 가져옵니다. [여기](https://www.cs.cmu.edu/~custom-diffusion/assets/data.zip)에서 데이터셋을 다운로드하고 압축을 풉니다. 직접 데이터셋을 사용하려면 [학습용 데이터셋 생성하기](create_dataset) 가이드를 참고하세요.

또한 'clip-retrieval'을 사용하여 200개의 실제 이미지를 수집하고, regularization으로서 이를 학습 데이터셋의 타겟 이미지와 결합합니다. 이렇게 하면 주어진 타겟 이미지에 대한 과적합을 방지할 수 있습니다. 다음 플래그를 사용하면 `prior_loss_weight=1.`로 `prior_preservation`, `real_prior` regularization을 활성화할 수 있습니다.
클래스_프롬프트`는 대상 이미지와 동일한 카테고리 이름이어야 합니다. 수집된 실제 이미지에는 `class_prompt`와 유사한 텍스트 캡션이 있습니다. 검색된 이미지는 `class_data_dir`에 저장됩니다. 생성된 이미지를 regularization으로 사용하기 위해 `real_prior`를 비활성화할 수 있습니다. 실제 이미지를 수집하려면 훈련 전에 이 명령을 먼저 사용하십시오.

```bash
pip install clip-retrieval
python retrieve.py --class_prompt cat --class_data_dir real_reg/samples_cat --num_class_images 200
```

**___참고: [stable-diffusion-2](https://huggingface.co/stabilityai/stable-diffusion-2) 768x768 모델을 사용하는 경우 '해상도'를 768로 변경하세요.___**

스크립트는 모델 체크포인트와 `pytorch_custom_diffusion_weights.bin` 파일을 생성하여 저장소에 저장합니다.

```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export OUTPUT_DIR="path-to-save-model"
export INSTANCE_DIR="./data/cat"

accelerate launch train_custom_diffusion.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --class_data_dir=./real_reg/samples_cat/ \
  --with_prior_preservation --real_prior --prior_loss_weight=1.0 \
  --class_prompt="cat" --num_class_images=200 \
  --instance_prompt="photo of a <new1> cat"  \
  --resolution=512  \
  --train_batch_size=2  \
  --learning_rate=1e-5  \
  --lr_warmup_steps=0 \
  --max_train_steps=250 \
  --scale_lr --hflip  \
  --modifier_token "<new1>" \
  --push_to_hub
```

**더 낮은 VRAM 요구 사항(GPU당 16GB)으로 더 빠르게 훈련하려면 `--enable_xformers_memory_efficient_attention`을 사용하세요. 설치 방법은 [가이드](https://github.com/facebookresearch/xformers)를 따르세요.**

가중치 및 편향(`wandb`)을 사용하여 실험을 추적하고 중간 결과를 저장하려면(강력히 권장합니다) 다음 단계를 따르세요:

* `wandb` 설치: `pip install wandb`.
* 로그인 : `wandb login`.
* 그런 다음 트레이닝을 시작하는 동안 `validation_prompt`를 지정하고 `report_to`를 `wandb`로 설정합니다. 다음과 같은 관련 인수를 구성할 수도 있습니다:
    * `num_validation_images`
    * `validation_steps`

```bash
accelerate launch train_custom_diffusion.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --class_data_dir=./real_reg/samples_cat/ \
  --with_prior_preservation --real_prior --prior_loss_weight=1.0 \
  --class_prompt="cat" --num_class_images=200 \
  --instance_prompt="photo of a <new1> cat"  \
  --resolution=512  \
  --train_batch_size=2  \
  --learning_rate=1e-5  \
  --lr_warmup_steps=0 \
  --max_train_steps=250 \
  --scale_lr --hflip  \
  --modifier_token "<new1>" \
  --validation_prompt="<new1> cat sitting in a bucket" \
  --report_to="wandb" \
  --push_to_hub
```

다음은 [Weights and Biases page](https://wandb.ai/sayakpaul/custom-diffusion/runs/26ghrcau)의 예시이며, 여러 학습 세부 정보와 함께 중간 결과들을 확인할 수 있습니다.

`--push_to_hub`를 지정하면 학습된 파라미터가 허깅 페이스 허브의 리포지토리에 푸시됩니다. 다음은 [예제 리포지토리](https://huggingface.co/sayakpaul/custom-diffusion-cat)입니다.

### 멀티 컨셉에 대한 학습 🐱🪵

[this](https://github.com/ShivamShrirao/diffusers/blob/main/examples/dreambooth/train_dreambooth.py)와 유사하게 각 컨셉에 대한 정보가 포함된 [json](https://github.com/adobe-research/custom-diffusion/blob/main/assets/concept_list.json) 파일을 제공합니다.

실제 이미지를 수집하려면 json 파일의 각 컨셉에 대해 이 명령을 실행합니다.

```bash
pip install clip-retrieval
python retrieve.py --class_prompt {} --class_data_dir {} --num_class_images 200
```

그럼 우리는 학습시킬 준비가 되었습니다!

```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export OUTPUT_DIR="path-to-save-model"

accelerate launch train_custom_diffusion.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --output_dir=$OUTPUT_DIR \
  --concepts_list=./concept_list.json \
  --with_prior_preservation --real_prior --prior_loss_weight=1.0 \
  --resolution=512  \
  --train_batch_size=2  \
  --learning_rate=1e-5  \
  --lr_warmup_steps=0 \
  --max_train_steps=500 \
  --num_class_images=200 \
  --scale_lr --hflip  \
  --modifier_token "<new1>+<new2>" \
  --push_to_hub
```

다음은 [Weights and Biases page](https://wandb.ai/sayakpaul/custom-diffusion/runs/3990tzkg)의 예시이며, 다른 학습 세부 정보와 함께 중간 결과들을 확인할 수 있습니다.

### 사람 얼굴에 대한 학습

사람 얼굴에 대한 파인튜닝을 위해 다음과 같은 설정이 더 효과적이라는 것을 확인했습니다: `learning_rate=5e-6`, `max_train_steps=1000 to 2000`, `freeze_model=crossattn`을 최소 15~20개의 이미지로 설정합니다.

실제 이미지를 수집하려면 훈련 전에 이 명령을 먼저 사용하십시오.

```bash
pip install clip-retrieval
python retrieve.py --class_prompt person --class_data_dir real_reg/samples_person --num_class_images 200
```

이제 학습을 시작하세요!

```bash
export MODEL_NAME="CompVis/stable-diffusion-v1-4"
export OUTPUT_DIR="path-to-save-model"
export INSTANCE_DIR="path-to-images"

accelerate launch train_custom_diffusion.py \
  --pretrained_model_name_or_path=$MODEL_NAME  \
  --instance_data_dir=$INSTANCE_DIR \
  --output_dir=$OUTPUT_DIR \
  --class_data_dir=./real_reg/samples_person/ \
  --with_prior_preservation --real_prior --prior_loss_weight=1.0 \
  --class_prompt="person" --num_class_images=200 \
  --instance_prompt="photo of a <new1> person"  \
  --resolution=512  \
  --train_batch_size=2  \
  --learning_rate=5e-6  \
  --lr_warmup_steps=0 \
  --max_train_steps=1000 \
  --scale_lr --hflip --noaug \
  --freeze_model crossattn \
  --modifier_token "<new1>" \
  --enable_xformers_memory_efficient_attention \
  --push_to_hub
```

## 추론

위 프롬프트를 사용하여 모델을 학습시킨 후에는 아래 프롬프트를 사용하여 추론을 실행할 수 있습니다. 프롬프트에 'modifier token'(예: 위 예제에서는 \<new1\>)을 반드시 포함해야 합니다.

```python
import torch
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", torch_dtype=torch.float16).to("cuda")
pipe.unet.load_attn_procs("path-to-save-model", weight_name="pytorch_custom_diffusion_weights.bin")
pipe.load_textual_inversion("path-to-save-model", weight_name="<new1>.bin")

image = pipe(
    "<new1> cat sitting in a bucket",
    num_inference_steps=100,
    guidance_scale=6.0,
    eta=1.0,
).images[0]
image.save("cat.png")
```

허브 리포지토리에서 이러한 매개변수를 직접 로드할 수 있습니다:

```python
import torch
from huggingface_hub.repocard import RepoCard
from diffusers import DiffusionPipeline

model_id = "sayakpaul/custom-diffusion-cat"
card = RepoCard.load(model_id)
base_model_id = card.data.to_dict()["base_model"]

pipe = DiffusionPipeline.from_pretrained(base_model_id, torch_dtype=torch.float16).to("cuda")
pipe.unet.load_attn_procs(model_id, weight_name="pytorch_custom_diffusion_weights.bin")
pipe.load_textual_inversion(model_id, weight_name="<new1>.bin")

image = pipe(
    "<new1> cat sitting in a bucket",
    num_inference_steps=100,
    guidance_scale=6.0,
    eta=1.0,
).images[0]
image.save("cat.png")
```

다음은 여러 컨셉으로 추론을 수행하는 예제입니다:

```python
import torch
from huggingface_hub.repocard import RepoCard
from diffusers import DiffusionPipeline

model_id = "sayakpaul/custom-diffusion-cat-wooden-pot"
card = RepoCard.load(model_id)
base_model_id = card.data.to_dict()["base_model"]

pipe = DiffusionPipeline.from_pretrained(base_model_id, torch_dtype=torch.float16).to("cuda")
pipe.unet.load_attn_procs(model_id, weight_name="pytorch_custom_diffusion_weights.bin")
pipe.load_textual_inversion(model_id, weight_name="<new1>.bin")
pipe.load_textual_inversion(model_id, weight_name="<new2>.bin")

image = pipe(
    "the <new1> cat sculpture in the style of a <new2> wooden pot",
    num_inference_steps=100,
    guidance_scale=6.0,
    eta=1.0,
).images[0]
image.save("multi-subject.png")
```

여기서 '고양이'와 '나무 냄비'는 여러 컨셉을 말합니다.

### 학습된 체크포인트에서 추론하기

`--checkpointing_steps`  인수를 사용한 경우 학습 과정에서 저장된 전체 체크포인트 중 하나에서 추론을 수행할 수도 있습니다.

## Grads를 None으로 설정

더 많은 메모리를 절약하려면 스크립트에 `--set_grads_to_none` 인수를 전달하세요. 이렇게 하면 성적이 0이 아닌 없음으로 설정됩니다. 그러나 특정 동작이 변경되므로 문제가 발생하면 이 인수를 제거하세요.

자세한 정보: https://pytorch.org/docs/stable/generated/torch.optim.Optimizer.zero_grad.html

## 실험 결과

실험에 대한 자세한 내용은 [당사 웹페이지](https://www.cs.cmu.edu/~custom-diffusion/)를 참조하세요.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/custom_diffusion.md" />

### 학습을 위한 데이터셋 만들기
https://huggingface.co/docs/diffusers/main/ko/training/create_dataset.md

# 학습을 위한 데이터셋 만들기

[Hub](https://huggingface.co/datasets?task_categories=task_categories:text-to-image&sort=downloads) 에는 모델 교육을 위한 많은 데이터셋이 있지만,
관심이 있거나 사용하고 싶은 데이터셋을 찾을 수 없는 경우 🤗 [Datasets](https://huggingface.co/docs/datasets) 라이브러리를 사용하여 데이터셋을 만들 수 있습니다.
데이터셋 구조는 모델을 학습하려는 작업에 따라 달라집니다.
가장 기본적인 데이터셋 구조는 unconditional 이미지 생성과 같은 작업을 위한 이미지 디렉토리입니다.
또 다른 데이터셋 구조는 이미지 디렉토리와 text-to-image 생성과 같은 작업에 해당하는 텍스트 캡션이 포함된 텍스트 파일일 수 있습니다.

이 가이드에는 파인 튜닝할 데이터셋을 만드는 두 가지 방법을 소개합니다:

- 이미지 폴더를 `--train_data_dir` 인수에 제공합니다.
- 데이터셋을 Hub에 업로드하고 데이터셋 리포지토리 id를 `--dataset_name` 인수에 전달합니다.

> [!TIP]
> 💡 학습에 사용할 이미지 데이터셋을 만드는 방법에 대한 자세한 내용은 [이미지 데이터셋 만들기](https://huggingface.co/docs/datasets/image_dataset) 가이드를 참고하세요.

## 폴더 형태로 데이터셋 구축하기

Unconditional 생성을 위해 이미지 폴더로 자신의 데이터셋을 구축할 수 있습니다.
학습 스크립트는 🤗 Datasets의 [ImageFolder](https://huggingface.co/docs/datasets/en/image_dataset#imagefolder) 빌더를 사용하여
자동으로 폴더에서 데이터셋을 구축합니다. 디렉토리 구조는 다음과 같아야 합니다 :

```bash
data_dir/xxx.png
data_dir/xxy.png
data_dir/[...]/xxz.png
```

데이터셋 디렉터리의 경로를 `--train_data_dir` 인수로 전달한 다음 학습을 시작할 수 있습니다:

```bash
accelerate launch train_unconditional.py \
    # argument로 폴더 지정하기 \
    --train_data_dir <path-to-train-directory> \
    <other-arguments>
```

## Hub에 데이터 올리기

> [!TIP]
> 💡 데이터셋을 만들고 Hub에 업로드하는 것에 대한 자세한 내용은 [🤗 Datasets을 사용한 이미지 검색](https://huggingface.co/blog/image-search-datasets) 게시물을 참고하세요.

PIL 인코딩된 이미지가 포함된 `이미지` 열을 생성하는 [이미지 폴더](https://huggingface.co/docs/datasets/image_load#imagefolder) 기능을 사용하여 데이터셋 생성을 시작합니다.

`data_dir` 또는 `data_files` 매개 변수를 사용하여 데이터셋의 위치를 지정할 수 있습니다.
`data_files` 매개변수는 특정 파일을 `train` 이나 `test` 로 분리한 데이터셋에 매핑하는 것을 지원합니다:

```python
from datasets import load_dataset

# 예시 1: 로컬 폴더
dataset = load_dataset("imagefolder", data_dir="path_to_your_folder")

# 예시 2: 로컬 파일 (지원 포맷 : tar, gzip, zip, xz, rar, zstd)
dataset = load_dataset("imagefolder", data_files="path_to_zip_file")

# 예시 3: 원격 파일 (지원 포맷 : tar, gzip, zip, xz, rar, zstd)
dataset = load_dataset(
    "imagefolder",
    data_files="https://download.microsoft.com/download/3/E/1/3E1C3F21-ECDB-4869-8368-6DEBA77B919F/kagglecatsanddogs_3367a.zip",
)

# 예시 4: 여러개로 분할
dataset = load_dataset(
    "imagefolder", data_files={"train": ["path/to/file1", "path/to/file2"], "test": ["path/to/file3", "path/to/file4"]}
)
```

[push_to_hub(https://huggingface.co/docs/datasets/v2.13.1/en/package_reference/main_classes#datasets.Dataset.push_to_hub) 을 사용해서 Hub에 데이터셋을 업로드 합니다:

```python
# 터미널에서 hf auth login 커맨드를 이미 실행했다고 가정합니다
dataset.push_to_hub("name_of_your_dataset")

# 개인 repo로 push 하고 싶다면, `private=True` 을 추가하세요:
dataset.push_to_hub("name_of_your_dataset", private=True)
```

이제 데이터셋 이름을 `--dataset_name` 인수에 전달하여 데이터셋을 학습에 사용할 수 있습니다:

```bash
accelerate launch --mixed_precision="fp16"  train_text_to_image.py \
  --pretrained_model_name_or_path="stable-diffusion-v1-5/stable-diffusion-v1-5" \
  --dataset_name="name_of_your_dataset" \
  <other-arguments>
```

## 다음 단계

데이터셋을 생성했으니 이제 학습 스크립트의 `train_data_dir` (데이터셋이 로컬이면) 혹은 `dataset_name` (Hub에 데이터셋을 올렸으면) 인수에 연결할 수 있습니다.

다음 단계에서는 데이터셋을 사용하여 [unconditional 생성](https://huggingface.co/docs/diffusers/v0.18.2/en/training/unconditional_training) 또는 [텍스트-이미지 생성](https://huggingface.co/docs/diffusers/training/text2image)을 위한 모델을 학습시켜보세요!


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/create_dataset.md" />

### InstructPix2Pix
https://huggingface.co/docs/diffusers/main/ko/training/instructpix2pix.md

# InstructPix2Pix

[InstructPix2Pix](https://huggingface.co/papers/2211.09800)는 text-conditioned diffusion 모델이 한 이미지에 편집을 따를 수 있도록 파인튜닝하는 방법입니다. 이 방법을 사용하여 파인튜닝된 모델은 다음을 입력으로 사용합니다:

<p align="center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/edit-instruction.png" alt="instructpix2pix-inputs" width=600/>
</p>

출력은 입력 이미지에 편집 지시가 반영된 "수정된" 이미지입니다:

<p align="center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/output-gs%407-igs%401-steps%4050.png" alt="instructpix2pix-output" width=600/>
</p>

`train_instruct_pix2pix.py` 스크립트([여기](https://github.com/huggingface/diffusers/blob/main/examples/instruct_pix2pix/train_instruct_pix2pix.py)에서 찾을 수 있습니다.)는 학습 절차를 설명하고 Stable Diffusion에 적용할 수 있는 방법을 보여줍니다.


*** `train_instruct_pix2pix.py`는 [원래 구현](https://github.com/timothybrooks/instruct-pix2pix)에 충실하면서 InstructPix2Pix 학습 절차를 구현하고 있지만, [소규모 데이터셋](https://huggingface.co/datasets/fusing/instructpix2pix-1000-samples)에서만 테스트를 했습니다. 이는 최종 결과에 영향을 끼칠 수 있습니다. 더 나은 결과를 위해, 더 큰 데이터셋에서 더 길게 학습하는 것을 권장합니다. [여기](https://huggingface.co/datasets/timbrooks/instructpix2pix-clip-filtered)에서 InstructPix2Pix 학습을 위해 큰 데이터셋을 찾을 수 있습니다.
***

## PyTorch로 로컬에서 실행하기

### 종속성(dependencies) 설치하기

이 스크립트를 실행하기 전에, 라이브러리의 학습 종속성을 설치하세요:

**중요**

최신 버전의 예제 스크립트를 성공적으로 실행하기 위해, **원본으로부터 설치**하는 것과 예제 스크립트를 자주 업데이트하고 예제별 요구사항을 설치하기 때문에 최신 상태로 유지하는 것을 권장합니다. 이를 위해, 새로운 가상 환경에서 다음 스텝을 실행하세요:

```bash
git clone https://github.com/huggingface/diffusers
cd diffusers
pip install -e .
```

cd 명령어로 예제 폴더로 이동하세요.
```bash
cd examples/instruct_pix2pix
```

이제 실행하세요.
```bash
pip install -r requirements.txt
```

그리고 [🤗Accelerate](https://github.com/huggingface/accelerate/) 환경에서 초기화하세요:

```bash
accelerate config
```

혹은 환경에 대한 질문 없이 기본적인 accelerate 구성을 사용하려면 다음을 실행하세요.

```bash
accelerate config default
```

혹은 사용 중인 환경이 notebook과 같은 대화형 쉘은 지원하지 않는 경우는 다음 절차를 따라주세요.

```python
from accelerate.utils import write_basic_config

write_basic_config()
```

### 예시

이전에 언급했듯이, 학습을 위해 [작은 데이터셋](https://huggingface.co/datasets/fusing/instructpix2pix-1000-samples)을 사용할 것입니다. 그 데이터셋은 InstructPix2Pix 논문에서 사용된 [원래의 데이터셋](https://huggingface.co/datasets/timbrooks/instructpix2pix-clip-filtered)보다 작은 버전입니다. 자신의 데이터셋을 사용하기 위해, [학습을 위한 데이터셋 만들기](create_dataset) 가이드를 참고하세요.

`MODEL_NAME` 환경 변수(허브 모델 레포지토리 또는 모델 가중치가 포함된 폴더 경로)를 지정하고 [`pretrained_model_name_or_path`](https://huggingface.co/docs/diffusers/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained.pretrained_model_name_or_path) 인수에 전달합니다. `DATASET_ID`에 데이터셋 이름을 지정해야 합니다:


```bash
export MODEL_NAME="stable-diffusion-v1-5/stable-diffusion-v1-5"
export DATASET_ID="fusing/instructpix2pix-1000-samples"
```

지금, 학습을 실행할 수 있습니다. 스크립트는 레포지토리의 하위 폴더의 모든 구성요소(`feature_extractor`, `scheduler`, `text_encoder`, `unet` 등)를 저장합니다.

```bash
accelerate launch --mixed_precision="fp16" train_instruct_pix2pix.py \
    --pretrained_model_name_or_path=$MODEL_NAME \
    --dataset_name=$DATASET_ID \
    --enable_xformers_memory_efficient_attention \
    --resolution=256 --random_flip \
    --train_batch_size=4 --gradient_accumulation_steps=4 --gradient_checkpointing \
    --max_train_steps=15000 \
    --checkpointing_steps=5000 --checkpoints_total_limit=1 \
    --learning_rate=5e-05 --max_grad_norm=1 --lr_warmup_steps=0 \
    --conditioning_dropout_prob=0.05 \
    --mixed_precision=fp16 \
    --seed=42 \
    --push_to_hub
```


추가적으로, 가중치와 바이어스를 학습 과정에 모니터링하여 검증 추론을 수행하는 것을 지원합니다. `report_to="wandb"`와 이 기능을 사용할 수 있습니다:

```bash
accelerate launch --mixed_precision="fp16" train_instruct_pix2pix.py \
    --pretrained_model_name_or_path=$MODEL_NAME \
    --dataset_name=$DATASET_ID \
    --enable_xformers_memory_efficient_attention \
    --resolution=256 --random_flip \
    --train_batch_size=4 --gradient_accumulation_steps=4 --gradient_checkpointing \
    --max_train_steps=15000 \
    --checkpointing_steps=5000 --checkpoints_total_limit=1 \
    --learning_rate=5e-05 --max_grad_norm=1 --lr_warmup_steps=0 \
    --conditioning_dropout_prob=0.05 \
    --mixed_precision=fp16 \
    --val_image_url="https://hf.co/datasets/diffusers/diffusers-images-docs/resolve/main/mountain.png" \
    --validation_prompt="make the mountains snowy" \
    --seed=42 \
    --report_to=wandb \
    --push_to_hub
 ```

모델 디버깅에 유용한 이 평가 방법 권장합니다. 이를 사용하기 위해 `wandb`를 설치하는 것을 주목해주세요. `pip install wandb`로 실행해 `wandb`를 설치할 수 있습니다.

[여기](https://wandb.ai/sayakpaul/instruct-pix2pix/runs/ctr3kovq), 몇 가지 평가 방법과 학습 파라미터를 포함하는 예시를 볼 수 있습니다.

 ***참고: 원본 논문에서, 저자들은 256x256 이미지 해상도로 학습한 모델로 512x512와 같은 더 큰 해상도로 잘 일반화되는 것을 볼 수 있었습니다. 이는 학습에 사용한 큰 데이터셋을 사용했기 때문입니다.***

 ## 다수의 GPU로 학습하기

`accelerate`는 원활한 다수의 GPU로 학습을 가능하게 합니다. `accelerate`로 분산 학습을 실행하는 [여기](https://huggingface.co/docs/accelerate/basic_tutorials/launch) 설명을 따라 해 주시기 바랍니다. 예시의 명령어 입니다:


```bash
accelerate launch --mixed_precision="fp16" --multi_gpu train_instruct_pix2pix.py \
 --pretrained_model_name_or_path=stable-diffusion-v1-5/stable-diffusion-v1-5 \
 --dataset_name=sayakpaul/instructpix2pix-1000-samples \
 --use_ema \
 --enable_xformers_memory_efficient_attention \
 --resolution=512 --random_flip \
 --train_batch_size=4 --gradient_accumulation_steps=4 --gradient_checkpointing \
 --max_train_steps=15000 \
 --checkpointing_steps=5000 --checkpoints_total_limit=1 \
 --learning_rate=5e-05 --lr_warmup_steps=0 \
 --conditioning_dropout_prob=0.05 \
 --mixed_precision=fp16 \
 --seed=42 \
 --push_to_hub
```

 ## 추론하기

일단 학습이 완료되면, 추론 할 수 있습니다:

 ```python
import PIL
import requests
import torch
from diffusers import StableDiffusionInstructPix2PixPipeline

model_id = "your_model_id"  # <- 이를 수정하세요.
pipe = StableDiffusionInstructPix2PixPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
generator = torch.Generator("cuda").manual_seed(0)

url = "https://huggingface.co/datasets/sayakpaul/sample-datasets/resolve/main/test_pix2pix_4.png"


def download_image(url):
    image = PIL.Image.open(requests.get(url, stream=True).raw)
    image = PIL.ImageOps.exif_transpose(image)
    image = image.convert("RGB")
    return image


image = download_image(url)
prompt = "wipe out the lake"
num_inference_steps = 20
image_guidance_scale = 1.5
guidance_scale = 10

edited_image = pipe(
    prompt,
    image=image,
    num_inference_steps=num_inference_steps,
    image_guidance_scale=image_guidance_scale,
    guidance_scale=guidance_scale,
    generator=generator,
).images[0]
edited_image.save("edited_image.png")
```

학습 스크립트를 사용해 얻은 예시의 모델 레포지토리는 여기 [sayakpaul/instruct-pix2pix](https://huggingface.co/sayakpaul/instruct-pix2pix)에서 확인할 수 있습니다.

성능을 위한 속도와 품질을 제어하기 위해 세 가지 파라미터를 사용하는 것이 좋습니다:

* `num_inference_steps`
* `image_guidance_scale`
* `guidance_scale`

특히, `image_guidance_scale`와 `guidance_scale`는 생성된("수정된") 이미지에서 큰 영향을 미칠 수 있습니다.([여기](https://twitter.com/RisingSayak/status/1628392199196151808?s=20)예시를 참고해주세요.)


만약 InstructPix2Pix 학습 방법을 사용해 몇 가지 흥미로운 방법을 찾고 있다면, 이 블로그 게시물[Instruction-tuning Stable Diffusion with InstructPix2Pix](https://huggingface.co/blog/instruction-tuning-sd)을 확인해주세요.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/instructpix2pix.md" />

### Textual-Inversion
https://huggingface.co/docs/diffusers/main/ko/training/text_inversion.md

# Textual-Inversion


[textual-inversion](https://huggingface.co/papers/2208.01618)은 소수의 예시 이미지에서 새로운 콘셉트를 포착하는 기법입니다. 이 기술은 원래 [Latent Diffusion](https://github.com/CompVis/latent-diffusion)에서 시연되었지만, 이후 [Stable Diffusion](https://huggingface.co/docs/diffusers/main/en/conceptual/stable_diffusion)과 같은 유사한 다른 모델에도 적용되었습니다. 학습된 콘셉트는 text-to-image 파이프라인에서 생성된 이미지를 더 잘 제어하는 데 사용할 수 있습니다. 이 모델은 텍스트 인코더의 임베딩 공간에서 새로운 '단어'를 학습하여 개인화된 이미지 생성을 위한 텍스트 프롬프트 내에서 사용됩니다.

![Textual Inversion example](https://textual-inversion.github.io/static/images/editing/colorful_teapot.JPG)
<small>By using just 3-5 images you can teach new concepts to a model such as Stable Diffusion for personalized image generation <a href="https://github.com/rinongal/textual_inversion">(image source)</a>.</small>

이 가이드에서는 textual-inversion으로 [`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) 모델을 학습하는 방법을 설명합니다. 이 가이드에서 사용된 모든 textual-inversion 학습 스크립트는 [여기](https://github.com/huggingface/diffusers/tree/main/examples/textual_inversion)에서 확인할 수 있습니다. 내부적으로 어떻게 작동하는지 자세히 살펴보고 싶으시다면 해당 링크를 참조해주시기 바랍니다.

> [!TIP]
> [Stable Diffusion Textual Inversion Concepts Library](https://huggingface.co/sd-concepts-library)에는 커뮤니티에서 제작한 학습된 textual-inversion 모델들이 있습니다. 시간이 지남에 따라 더 많은 콘셉트들이 추가되어 유용한 리소스로 성장할 것입니다!

시작하기 전에 학습을 위한 의존성 라이브러리들을 설치해야 합니다:

```bash
pip install diffusers accelerate transformers
```

의존성 라이브러리들의 설치가 완료되면, [🤗Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화시킵니다.

```bash
accelerate config
```

별도의 설정없이, 기본 🤗Accelerate 환경을 설정하려면 다음과 같이 하세요:

```bash
accelerate config default
```

또는 사용 중인 환경이 노트북과 같은 대화형 셸을 지원하지 않는다면, 다음과 같이 사용할 수 있습니다:

```py
from accelerate.utils import write_basic_config

write_basic_config()
```

마지막으로, Memory-Efficient Attention을 통해 메모리 사용량을 줄이기 위해 [xFormers](https://huggingface.co/docs/diffusers/main/en/training/optimization/xformers)를 설치합니다. xFormers를 설치한 후, 학습 스크립트에 `--enable_xformers_memory_efficient_attention` 인자를 추가합니다. xFormers는 Flax에서 지원되지 않습니다.

## 허브에 모델 업로드하기

모델을 허브에 저장하려면, 학습 스크립트에 다음 인자를 추가해야 합니다.

```bash
--push_to_hub
```

## 체크포인트 저장 및 불러오기

학습중에 모델의 체크포인트를 정기적으로 저장하는 것이 좋습니다. 이렇게 하면 어떤 이유로든 학습이 중단된 경우 저장된 체크포인트에서 학습을 다시 시작할 수 있습니다. 학습 스크립트에 다음 인자를 전달하면 500단계마다 전체 학습 상태가 `output_dir`의 하위 폴더에 체크포인트로서 저장됩니다.

```bash
--checkpointing_steps=500
```

저장된 체크포인트에서 학습을 재개하려면, 학습 스크립트와 재개할 특정 체크포인트에 다음 인자를 전달하세요.

```bash
--resume_from_checkpoint="checkpoint-1500"
```

## 파인 튜닝

학습용 데이터셋으로 [고양이 장난감 데이터셋](https://huggingface.co/datasets/diffusers/cat_toy_example)을 다운로드하여 디렉토리에 저장하세요. 여러분만의 고유한 데이터셋을 사용하고자 한다면, [학습용 데이터셋 만들기](https://huggingface.co/docs/diffusers/training/create_dataset) 가이드를 살펴보시기 바랍니다.

```py
from huggingface_hub import snapshot_download

local_dir = "./cat"
snapshot_download(
    "diffusers/cat_toy_example", local_dir=local_dir, repo_type="dataset", ignore_patterns=".gitattributes"
)
```

모델의 리포지토리 ID(또는 모델 가중치가 포함된 디렉터리 경로)를 `MODEL_NAME` 환경 변수에 할당하고, 해당 값을 [`pretrained_model_name_or_path`](https://huggingface.co/docs/diffusers/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained.pretrained_model_name_or_path) 인자에 전달합니다. 그리고 이미지가 포함된 디렉터리 경로를 `DATA_DIR` 환경 변수에 할당합니다.

이제 [학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion.py)를 실행할 수 있습니다. 스크립트는 다음 파일을 생성하고 리포지토리에 저장합니다.

- `learned_embeds.bin`
- `token_identifier.txt`
- `type_of_concept.txt`.

> [!TIP]
> 💡V100 GPU 1개를 기준으로 전체 학습에는 최대 1시간이 걸립니다. 학습이 완료되기를 기다리는 동안 궁금한 점이 있으면 아래 섹션에서 [textual-inversion이 어떻게 작동하는지](https://huggingface.co/docs/diffusers/training/text_inversion#how-it-works) 자유롭게 확인하세요 !

<frameworkcontent>
<pt>
```bash
export MODEL_NAME="stable-diffusion-v1-5/stable-diffusion-v1-5"
export DATA_DIR="./cat"

accelerate launch textual_inversion.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATA_DIR \
  --learnable_property="object" \
  --placeholder_token="<cat-toy>" --initializer_token="toy" \
  --resolution=512 \
  --train_batch_size=1 \
  --gradient_accumulation_steps=4 \
  --max_train_steps=3000 \
  --learning_rate=5.0e-04 --scale_lr \
  --lr_scheduler="constant" \
  --lr_warmup_steps=0 \
  --output_dir="textual_inversion_cat" \
  --push_to_hub
```

> [!TIP]
> 💡학습 성능을 올리기 위해, 플레이스홀더 토큰(`<cat-toy>`)을 (단일한 임베딩 벡터가 아닌) 복수의 임베딩 벡터로 표현하는 것 역시 고려할 있습니다.  이러한 트릭이 모델이 보다 복잡한 이미지의 스타일(앞서 말한 콘셉트)을 더 잘 캡처하는 데 도움이 될 수 있습니다. 복수의 임베딩 벡터 학습을 활성화하려면 다음 옵션을 전달하십시오.
>
> ```bash
> --num_vectors=5
> ```
</pt>
<jax>

TPU에 액세스할 수 있는 경우, [Flax 학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion_flax.py)를 사용하여 더 빠르게 모델을 학습시켜보세요. (물론 GPU에서도 작동합니다.) 동일한 설정에서 Flax 학습 스크립트는 PyTorch 학습 스크립트보다 최소 70% 더 빨라야 합니다! ⚡️

시작하기 앞서 Flax에 대한 의존성 라이브러리들을 설치해야 합니다.

```bash
pip install -U -r requirements_flax.txt
```

모델의 리포지토리 ID(또는 모델 가중치가 포함된 디렉터리 경로)를 `MODEL_NAME` 환경 변수에 할당하고, 해당 값을 [`pretrained_model_name_or_path`](https://huggingface.co/docs/diffusers/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained.pretrained_model_name_or_path) 인자에 전달합니다.

그런 다음 [학습 스크립트](https://github.com/huggingface/diffusers/blob/main/examples/textual_inversion/textual_inversion_flax.py)를 시작할 수 있습니다.

```bash
export MODEL_NAME="duongna/stable-diffusion-v1-4-flax"
export DATA_DIR="./cat"

python textual_inversion_flax.py \
  --pretrained_model_name_or_path=$MODEL_NAME \
  --train_data_dir=$DATA_DIR \
  --learnable_property="object" \
  --placeholder_token="<cat-toy>" --initializer_token="toy" \
  --resolution=512 \
  --train_batch_size=1 \
  --max_train_steps=3000 \
  --learning_rate=5.0e-04 --scale_lr \
  --output_dir="textual_inversion_cat" \
  --push_to_hub
```
</jax>
</frameworkcontent>

### 중간 로깅

모델의 학습 진행 상황을 추적하는 데 관심이 있는 경우, 학습 과정에서 생성된 이미지를 저장할 수 있습니다. 학습 스크립트에 다음 인수를 추가하여 중간 로깅을 활성화합니다.

- `validation_prompt` : 샘플을 생성하는 데 사용되는 프롬프트(기본값은 `None`으로 설정되며, 이 때 중간 로깅은 비활성화됨)
- `num_validation_images` : 생성할 샘플 이미지 수
- `validation_steps` : `validation_prompt`로부터 샘플 이미지를 생성하기 전 스텝의 수

```bash
--validation_prompt="A <cat-toy> backpack"
--num_validation_images=4
--validation_steps=100
```

## 추론

모델을 학습한 후에는, 해당 모델을 `StableDiffusionPipeline`을 사용하여 추론에 사용할 수 있습니다.

textual-inversion 스크립트는 기본적으로 textual-inversion을 통해 얻어진 임베딩 벡터만을 저장합니다. 해당 임베딩 벡터들은 텍스트 인코더의 임베딩 행렬에 추가되어 있습습니다.

<frameworkcontent>
<pt>
> [!TIP]
> 💡 커뮤니티는 [sd-concepts-library](https://huggingface.co/sd-concepts-library) 라는 대규모의 textual-inversion 임베딩 벡터 라이브러리를 만들었습니다. textual-inversion 임베딩을 밑바닥부터 학습하는 대신, 해당 라이브러리에 본인이 찾는 textual-inversion 임베딩이 이미 추가되어 있지 않은지를 확인하는 것도 좋은 방법이 될 것 같습니다.

textual-inversion 임베딩 벡터을 불러오기 위해서는, 먼저 해당 임베딩 벡터를 학습할 때 사용한 모델을 불러와야 합니다. 여기서는  [`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/docs/diffusers/training/stable-diffusion-v1-5/stable-diffusion-v1-5) 모델이 사용되었다고 가정하고 불러오겠습니다.

```python
from diffusers import StableDiffusionPipeline
import torch

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16).to("cuda")
```

다음으로 `TextualInversionLoaderMixin.load_textual_inversion` 함수를 통해, textual-inversion 임베딩 벡터를 불러와야 합니다. 여기서 우리는 이전의 `<cat-toy>` 예제의 임베딩을 불러올 것입니다.

```python
pipe.load_textual_inversion("sd-concepts-library/cat-toy")
```

이제 플레이스홀더 토큰(`<cat-toy>`)이 잘 동작하는지를 확인하는 파이프라인을 실행할 수 있습니다.

```python
prompt = "A <cat-toy> backpack"

image = pipe(prompt, num_inference_steps=50).images[0]
image.save("cat-backpack.png")
```

`TextualInversionLoaderMixin.load_textual_inversion`은 Diffusers 형식으로 저장된 텍스트 임베딩 벡터를 로드할 수 있을 뿐만 아니라, [Automatic1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui) 형식으로 저장된 임베딩 벡터도 로드할 수 있습니다. 이렇게 하려면, 먼저 [civitAI](https://civitai.com/models/3036?modelVersionId=8387)에서 임베딩 벡터를 다운로드한 다음 로컬에서 불러와야 합니다.

```python
pipe.load_textual_inversion("./charturnerv2.pt")
```
</pt>
<jax>

현재 Flax에 대한 `load_textual_inversion` 함수는 없습니다. 따라서 학습 후 textual-inversion 임베딩 벡터가 모델의 일부로서 저장되었는지를 확인해야 합니다. 그런 다음은 다른 Flax 모델과 마찬가지로 실행할 수 있습니다.

```python
import jax
import numpy as np
from flax.jax_utils import replicate
from flax.training.common_utils import shard
from diffusers import FlaxStableDiffusionPipeline

model_path = "path-to-your-trained-model"
pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(model_path, dtype=jax.numpy.bfloat16)

prompt = "A <cat-toy> backpack"
prng_seed = jax.random.PRNGKey(0)
num_inference_steps = 50

num_samples = jax.device_count()
prompt = num_samples * [prompt]
prompt_ids = pipeline.prepare_inputs(prompt)

# shard inputs and rng
params = replicate(params)
prng_seed = jax.random.split(prng_seed, jax.device_count())
prompt_ids = shard(prompt_ids)

images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
image.save("cat-backpack.png")
```
</jax>
</frameworkcontent>

## 작동 방식

![Diagram from the paper showing overview](https://textual-inversion.github.io/static/images/training/training.JPG)
<small>Architecture overview from the Textual Inversion <a href="https://textual-inversion.github.io/">blog post.</a></small>

일반적으로 텍스트 프롬프트는 모델에 전달되기 전에 임베딩으로 토큰화됩니다. textual-inversion은 비슷한 작업을 수행하지만, 위 다이어그램의 특수 토큰 `S*`로부터 새로운 토큰 임베딩 `v*`를 학습합니다. 모델의 아웃풋은 디퓨전 모델을 조정하는 데 사용되며, 디퓨전 모델이 단 몇 개의 예제 이미지에서 신속하고 새로운 콘셉트를 이해하는 데 도움을 줍니다.

이를 위해 textual-inversion은 제너레이터 모델과 학습용 이미지의 노이즈 버전을 사용합니다. 제너레이터는 노이즈가 적은 버전의 이미지를 예측하려고 시도하며 토큰 임베딩 `v*`은 제너레이터의 성능에 따라 최적화됩니다. 토큰 임베딩이 새로운 콘셉트를 성공적으로 포착하면 디퓨전 모델에 더 유용한 정보를 제공하고 노이즈가 적은 더 선명한 이미지를 생성하는 데 도움이 됩니다. 이러한 최적화 프로세스는 일반적으로 다양한 프롬프트와 이미지에 수천 번에 노출됨으로써 이루어집니다.



<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/text_inversion.md" />

### ControlNet
https://huggingface.co/docs/diffusers/main/ko/training/controlnet.md

# ControlNet

[Adding Conditional Control to Text-to-Image Diffusion Models](https://huggingface.co/papers/2302.05543) (ControlNet)은 Lvmin Zhang과 Maneesh Agrawala에 의해 쓰여졌습니다.

이 예시는 [원본 ControlNet 리포지토리에서 예시 학습하기](https://github.com/lllyasviel/ControlNet/blob/main/docs/train.md)에 기반합니다. ControlNet은 원들을 채우기 위해 [small synthetic dataset](https://huggingface.co/datasets/fusing/fill50k)을 사용해서 학습됩니다.

## 의존성 설치하기

아래의 스크립트를 실행하기 전에, 라이브러리의 학습 의존성을 설치해야 합니다.

> [!WARNING]
> 가장 최신 버전의 예시 스크립트를 성공적으로 실행하기 위해서는, 소스에서 설치하고 최신 버전의 설치를 유지하는 것을 강력하게 추천합니다. 우리는 예시 스크립트들을 자주 업데이트하고 예시에 맞춘 특정한 요구사항을 설치합니다.

위 사항을 만족시키기 위해서, 새로운 가상환경에서 다음 일련의 스텝을 실행하세요:

```bash
git clone https://github.com/huggingface/diffusers
cd diffusers
pip install -e .
```

그 다음에는 [예시 폴더](https://github.com/huggingface/diffusers/tree/main/examples/controlnet)으로 이동합니다.

```bash
cd examples/controlnet
```

이제 실행하세요:

```bash
pip install -r requirements.txt
```

[🤗Accelerate](https://github.com/huggingface/accelerate/) 환경을 초기화 합니다:

```bash
accelerate config
```

혹은 여러분의 환경이 무엇인지 몰라도 기본적인 🤗Accelerate 구성으로 초기화할 수 있습니다:

```bash
accelerate config default
```

혹은 당신의 환경이 노트북 같은 상호작용하는 쉘을 지원하지 않는다면, 아래의 코드로 초기화 할 수 있습니다:

```python
from accelerate.utils import write_basic_config

write_basic_config()
```

자체 데이터셋을 사용하기 위해서는 [학습을 위한 데이터셋 생성하기](create_dataset) 가이드를 확인하세요.

## 학습

이 학습에 사용될 다음 이미지들을 다운로드하세요:

```sh
wget https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_1.png

wget https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_2.png
```

`MODEL_NAME` 환경 변수 (Hub 모델 리포지토리 아이디 혹은 모델 가중치가 있는 디렉토리로 가는 주소)를 명시하고 [`pretrained_model_name_or_path`](https://huggingface.co/docs/diffusers/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained.pretrained_model_name_or_path) 인자로 환경변수를 보냅니다.

학습 스크립트는 당신의 리포지토리에 `diffusion_pytorch_model.bin` 파일을 생성하고 저장합니다.

```bash
export MODEL_DIR="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="path to save model"

accelerate launch train_controlnet.py \
 --pretrained_model_name_or_path=$MODEL_DIR \
 --output_dir=$OUTPUT_DIR \
 --dataset_name=fusing/fill50k \
 --resolution=512 \
 --learning_rate=1e-5 \
 --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
 --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
 --train_batch_size=4 \
 --push_to_hub
```

이 기본적인 설정으로는 ~38GB VRAM이 필요합니다.

기본적으로 학습 스크립트는 결과를 텐서보드에 기록합니다. 가중치(weight)와 편향(bias)을 사용하기 위해 `--report_to wandb` 를 전달합니다.

더 작은 batch(배치) 크기로 gradient accumulation(기울기 누적)을 하면 학습 요구사항을 ~20 GB VRAM으로 줄일 수 있습니다.

```bash
export MODEL_DIR="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="path to save model"

accelerate launch train_controlnet.py \
 --pretrained_model_name_or_path=$MODEL_DIR \
 --output_dir=$OUTPUT_DIR \
 --dataset_name=fusing/fill50k \
 --resolution=512 \
 --learning_rate=1e-5 \
 --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
 --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
 --train_batch_size=1 \
 --gradient_accumulation_steps=4 \
  --push_to_hub
```

## 여러개 GPU로 학습하기

`accelerate` 은 seamless multi-GPU 학습을 고려합니다. `accelerate`과 함께 분산된 학습을 실행하기 위해 [여기](https://huggingface.co/docs/accelerate/basic_tutorials/launch)
의 설명을 확인하세요. 아래는 예시 명령어입니다:

```bash
export MODEL_DIR="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="path to save model"

accelerate launch --mixed_precision="fp16" --multi_gpu train_controlnet.py \
 --pretrained_model_name_or_path=$MODEL_DIR \
 --output_dir=$OUTPUT_DIR \
 --dataset_name=fusing/fill50k \
 --resolution=512 \
 --learning_rate=1e-5 \
 --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
 --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
 --train_batch_size=4 \
 --mixed_precision="fp16" \
 --tracker_project_name="controlnet-demo" \
 --report_to=wandb \
  --push_to_hub
```

## 예시 결과

#### 배치 사이즈 8로 300 스텝 이후:

| |  |
|-------------------|:-------------------------:|
| | 푸른 배경과 빨간 원  |
![conditioning image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_1.png) | ![푸른 배경과 빨간 원](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/red_circle_with_blue_background_300_steps.png) |
| | 갈색 꽃 배경과 청록색 원 |
![conditioning image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_2.png) | ![갈색 꽃 배경과 청록색 원](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/cyan_circle_with_brown_floral_background_300_steps.png) |

#### 배치 사이즈 8로 6000 스텝 이후:

| |  |
|-------------------|:-------------------------:|
| | 푸른 배경과 빨간 원  |
![conditioning image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_1.png) | ![푸른 배경과 빨간 원](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/red_circle_with_blue_background_6000_steps.png) |
| | 갈색 꽃 배경과 청록색 원 |
![conditioning image](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/conditioning_image_2.png) | ![갈색 꽃 배경과 청록색 원](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/controlnet_training/cyan_circle_with_brown_floral_background_6000_steps.png) |

## 16GB GPU에서 학습하기

16GB GPU에서 학습하기 위해 다음의 최적화를 진행하세요:

- 기울기 체크포인트 저장하기
- bitsandbyte의 [8-bit optimizer](https://github.com/TimDettmers/bitsandbytes#requirements--installation)가 설치되지 않았다면 링크에 연결된 설명서를 보세요.

이제 학습 스크립트를 시작할 수 있습니다:

```bash
export MODEL_DIR="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="path to save model"

accelerate launch train_controlnet.py \
 --pretrained_model_name_or_path=$MODEL_DIR \
 --output_dir=$OUTPUT_DIR \
 --dataset_name=fusing/fill50k \
 --resolution=512 \
 --learning_rate=1e-5 \
 --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
 --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
 --train_batch_size=1 \
 --gradient_accumulation_steps=4 \
 --gradient_checkpointing \
 --use_8bit_adam \
  --push_to_hub
```

## 12GB GPU에서 학습하기

12GB GPU에서 실행하기 위해 다음의 최적화를 진행하세요:

- 기울기 체크포인트 저장하기
- bitsandbyte의 8-bit [optimizer](https://github.com/TimDettmers/bitsandbytes#requirements--installation)(가 설치되지 않았다면 링크에 연결된 설명서를 보세요)
- [xFormers](https://huggingface.co/docs/diffusers/training/optimization/xformers)(가 설치되지 않았다면 링크에 연결된 설명서를 보세요)
- 기울기를 `None`으로 설정

```bash
export MODEL_DIR="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="path to save model"

accelerate launch train_controlnet.py \
 --pretrained_model_name_or_path=$MODEL_DIR \
 --output_dir=$OUTPUT_DIR \
 --dataset_name=fusing/fill50k \
 --resolution=512 \
 --learning_rate=1e-5 \
 --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
 --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
 --train_batch_size=1 \
 --gradient_accumulation_steps=4 \
 --gradient_checkpointing \
 --use_8bit_adam \
 --enable_xformers_memory_efficient_attention \
 --set_grads_to_none \
  --push_to_hub
```

`pip install xformers`으로 `xformers`을 확실히 설치하고 `enable_xformers_memory_efficient_attention`을 사용하세요.

## 8GB GPU에서 학습하기

우리는 ControlNet을 지원하기 위한 DeepSpeed를 철저하게 테스트하지 않았습니다. 환경설정이 메모리를 저장할 때,
그 환경이 성공적으로 학습했는지를 확정하지 않았습니다. 성공한 학습 실행을 위해 설정을 변경해야 할 가능성이 높습니다.

8GB GPU에서 실행하기 위해 다음의 최적화를 진행하세요:

- 기울기 체크포인트 저장하기
- bitsandbyte의 8-bit [optimizer](https://github.com/TimDettmers/bitsandbytes#requirements--installation)(가 설치되지 않았다면 링크에 연결된 설명서를 보세요)
- [xFormers](https://huggingface.co/docs/diffusers/training/optimization/xformers)(가 설치되지 않았다면 링크에 연결된 설명서를 보세요)
- 기울기를 `None`으로 설정
- DeepSpeed stage 2 변수와 optimizer 없에기
- fp16 혼합 정밀도(precision)

[DeepSpeed](https://www.deepspeed.ai/)는 CPU 또는 NVME로 텐서를 VRAM에서 오프로드할 수 있습니다.
이를 위해서 훨씬 더 많은 RAM(약 25 GB)가 필요합니다.

DeepSpeed stage 2를 활성화하기 위해서 `accelerate config`로 환경을 구성해야합니다.

구성(configuration) 파일은 이런 모습이어야 합니다:

```yaml
compute_environment: LOCAL_MACHINE
deepspeed_config:
  gradient_accumulation_steps: 4
  offload_optimizer_device: cpu
  offload_param_device: cpu
  zero3_init_flag: false
  zero_stage: 2
distributed_type: DEEPSPEED
```

<팁>

[문서](https://huggingface.co/docs/accelerate/usage_guides/deepspeed)를 더 많은 DeepSpeed 설정 옵션을 위해 보세요.

<팁>

기본 Adam optimizer를 DeepSpeed'의 Adam
`deepspeed.ops.adam.DeepSpeedCPUAdam` 으로 바꾸면 상당한 속도 향상을 이룰수 있지만,
Pytorch와 같은 버전의 CUDA toolchain이 필요합니다. 8-비트 optimizer는 현재 DeepSpeed와
호환되지 않는 것 같습니다.

```bash
export MODEL_DIR="stable-diffusion-v1-5/stable-diffusion-v1-5"
export OUTPUT_DIR="path to save model"

accelerate launch train_controlnet.py \
 --pretrained_model_name_or_path=$MODEL_DIR \
 --output_dir=$OUTPUT_DIR \
 --dataset_name=fusing/fill50k \
 --resolution=512 \
 --validation_image "./conditioning_image_1.png" "./conditioning_image_2.png" \
 --validation_prompt "red circle with blue background" "cyan circle with brown floral background" \
 --train_batch_size=1 \
 --gradient_accumulation_steps=4 \
 --gradient_checkpointing \
 --enable_xformers_memory_efficient_attention \
 --set_grads_to_none \
 --mixed_precision fp16 \
 --push_to_hub
```

## 추론

학습된 모델은 `StableDiffusionControlNetPipeline`과 함께 실행될 수 있습니다.
`base_model_path`와 `controlnet_path` 에 값을 지정하세요 `--pretrained_model_name_or_path` 와
`--output_dir` 는 학습 스크립트에 개별적으로 지정됩니다.

```py
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
from diffusers.utils import load_image
import torch

base_model_path = "path to model"
controlnet_path = "path to controlnet"

controlnet = ControlNetModel.from_pretrained(controlnet_path, torch_dtype=torch.float16)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    base_model_path, controlnet=controlnet, torch_dtype=torch.float16
)

# 더 빠른 스케줄러와 메모리 최적화로 diffusion 프로세스 속도 올리기
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
# xformers가 설치되지 않으면 아래 줄을 삭제하기
pipe.enable_xformers_memory_efficient_attention()

pipe.enable_model_cpu_offload()

control_image = load_image("./conditioning_image_1.png")
prompt = "pale golden rod circle with old lace background"

# 이미지 생성하기
generator = torch.manual_seed(0)
image = pipe(prompt, num_inference_steps=20, generator=generator, image=control_image).images[0]

image.save("./output.png")
```


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/controlnet.md" />

### 새로운 작업에 대한 모델을 적용하기
https://huggingface.co/docs/diffusers/main/ko/training/adapt_a_model.md

# 새로운 작업에 대한 모델을 적용하기

많은 diffusion 시스템은 같은 구성 요소들을 공유하므로 한 작업에 대해 사전학습된 모델을 완전히 다른 작업에 적용할 수 있습니다.

이 인페인팅을 위한 가이드는 사전학습된 `UNet2DConditionModel`의 아키텍처를 초기화하고 수정하여 사전학습된 text-to-image 모델을 어떻게 인페인팅에 적용하는지를 알려줄 것입니다.

## UNet2DConditionModel 파라미터 구성

`UNet2DConditionModel`은 [input sample](https://huggingface.co/docs/diffusers/v0.16.0/en/api/models#diffusers.UNet2DConditionModel.in_channels)에서 4개의 채널을 기본적으로 허용합니다. 예를 들어,  [`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)와 같은 사전학습된 text-to-image 모델을 불러오고 `in_channels`의 수를 확인합니다:

```py
from diffusers import StableDiffusionPipeline

pipeline = StableDiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5")
pipeline.unet.config["in_channels"]
4
```

인페인팅은 입력 샘플에 9개의 채널이 필요합니다. [`runwayml/stable-diffusion-inpainting`](https://huggingface.co/runwayml/stable-diffusion-inpainting)와 같은 사전학습된 인페인팅 모델에서 이 값을 확인할 수 있습니다:

```py
from diffusers import StableDiffusionPipeline

pipeline = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-inpainting")
pipeline.unet.config["in_channels"]
9
```

인페인팅에 대한 text-to-image 모델을 적용하기 위해, `in_channels` 수를 4에서 9로 수정해야 할 것입니다.

사전학습된 text-to-image 모델의 가중치와 `UNet2DConditionModel`을 초기화하고 `in_channels`를 9로 수정해 주세요. `in_channels`의 수를 수정하면 크기가 달라지기 때문에 크기가 안 맞는 오류를 피하기 위해 `ignore_mismatched_sizes=True` 및 `low_cpu_mem_usage=False`를 설정해야 합니다.

```py
from diffusers import UNet2DConditionModel

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
unet = UNet2DConditionModel.from_pretrained(
    model_id, subfolder="unet", in_channels=9, low_cpu_mem_usage=False, ignore_mismatched_sizes=True
)
```

Text-to-image 모델로부터 다른 구성 요소의 사전학습된 가중치는 체크포인트로부터 초기화되지만 `unet`의 입력 채널 가중치 (`conv_in.weight`)는 랜덤하게 초기화됩니다. 그렇지 않으면 모델이 노이즈를 리턴하기 때문에 인페인팅의 모델을 파인튜닝 할 때 중요합니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/training/adapt_a_model.md" />

### 스케줄러
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/schedulers.md

# 스케줄러

diffusion 파이프라인은 diffusion 모델, 스케줄러 등의 컴포넌트들로 구성됩니다. 그리고 파이프라인 안의 일부 컴포넌트를 다른 컴포넌트로 교체하는 식의 커스터마이징 역시 가능합니다.  이와 같은 컴포넌트 커스터마이징의 가장 대표적인 예시가 바로 [스케줄러](../api/schedulers/overview.md)를 교체하는 것입니다.



스케쥴러는 다음과 같이 diffusion 시스템의 전반적인 디노이징 프로세스를 정의합니다.

- 디노이징 스텝을 얼마나 가져가야 할까?
- 확률적으로(stochastic) 혹은 확정적으로(deterministic)?
- 디노이징 된 샘플을 찾아내기 위해 어떤 알고리즘을 사용해야 할까?

이러한 프로세스는 다소 난해하고, 디노이징 속도와 디노이징 퀄리티 사이의 트레이드 오프를 정의해야 하는 문제가 될 수 있습니다. 주어진 파이프라인에 어떤 스케줄러가 가장 적합한지를 정량적으로 판단하는 것은 매우 어려운 일입니다. 이로 인해 일단 해당 스케줄러를 직접 사용하여, 생성되는 이미지를 직접 눈으로 보며, 정성적으로 성능을 판단해보는 것이 추천되곤 합니다.





## 파이프라인 불러오기

먼저 스테이블 diffusion 파이프라인을 불러오도록 해보겠습니다. 물론 스테이블 diffusion을 사용하기 위해서는, 허깅페이스 허브에 등록된 사용자여야 하며, 관련 [라이센스](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)에 동의해야 한다는 점을 잊지 말아주세요.

*역자 주: 다만, 현재 신규로 생성한 허깅페이스 계정에 대해서는 라이센스 동의를 요구하지 않는 것으로 보입니다!*

```python
from huggingface_hub import login
from diffusers import DiffusionPipeline
import torch

# first we need to login with our access token
login()

# Now we can download the pipeline
pipeline = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16)
```

다음으로, GPU로 이동합니다.

```python
pipeline.to("cuda")
```





## 스케줄러 액세스

스케줄러는 언제나 파이프라인의 컴포넌트로서 존재하며, 일반적으로 파이프라인 인스턴스 내에 `scheduler`라는 이름의 속성(property)으로 정의되어 있습니다.

```python
pipeline.scheduler
```

**Output**:

```
PNDMScheduler {
  "_class_name": "PNDMScheduler",
  "_diffusers_version": "0.8.0.dev0",
  "beta_end": 0.012,
  "beta_schedule": "scaled_linear",
  "beta_start": 0.00085,
  "clip_sample": false,
  "num_train_timesteps": 1000,
  "set_alpha_to_one": false,
  "skip_prk_steps": true,
  "steps_offset": 1,
  "trained_betas": null
}
```

출력 결과를 통해, 우리는 해당 스케줄러가 `PNDMScheduler`의 인스턴스라는 것을 알 수 있습니다. 이제 `PNDMScheduler`와 다른 스케줄러들의 성능을 비교해보도록 하겠습니다. 먼저 테스트에 사용할 프롬프트를 다음과 같이 정의해보도록 하겠습니다.

```python
prompt = "A photograph of an astronaut riding a horse on Mars, high resolution, high definition."
```

다음으로 유사한 이미지 생성을 보장하기 위해서, 다음과 같이 랜덤시드를 고정해주도록 하겠습니다.

```python
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator).images[0]
image
```

<p align="center">
    <br>
    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_pndm.png" width="400"/>
    <br>
</p>




## 스케줄러 교체하기

다음으로 파이프라인의 스케줄러를 다른 스케줄러로 교체하는 방법에 대해 알아보겠습니다. 모든 스케줄러는 `SchedulerMixin.compatibles`라는 속성(property)을 갖고 있습니다. 해당 속성은 **호환 가능한** 스케줄러들에 대한 정보를 담고 있습니다.

```python
pipeline.scheduler.compatibles
```

**Output**:

```
[diffusers.schedulers.scheduling_lms_discrete.LMSDiscreteScheduler,
 diffusers.schedulers.scheduling_ddim.DDIMScheduler,
 diffusers.schedulers.scheduling_dpmsolver_multistep.DPMSolverMultistepScheduler,
 diffusers.schedulers.scheduling_euler_discrete.EulerDiscreteScheduler,
 diffusers.schedulers.scheduling_pndm.PNDMScheduler,
 diffusers.schedulers.scheduling_ddpm.DDPMScheduler,
 diffusers.schedulers.scheduling_euler_ancestral_discrete.EulerAncestralDiscreteScheduler]
```

호환되는 스케줄러들을 살펴보면 아래와 같습니다.

- `LMSDiscreteScheduler`,
- `DDIMScheduler`,
- `DPMSolverMultistepScheduler`,
- `EulerDiscreteScheduler`,
- `PNDMScheduler`,
- `DDPMScheduler`,
- `EulerAncestralDiscreteScheduler`.

앞서 정의했던 프롬프트를 사용해서 각각의 스케줄러들을 비교해보도록 하겠습니다.

먼저 파이프라인 안의 스케줄러를 바꾸기 위해 `ConfigMixin.config` 속성과 `ConfigMixin.from_config()` 메서드를 활용해보려고 합니다.



```python
pipeline.scheduler.config
```

**Output**:

```
FrozenDict([('num_train_timesteps', 1000),
            ('beta_start', 0.00085),
            ('beta_end', 0.012),
            ('beta_schedule', 'scaled_linear'),
            ('trained_betas', None),
            ('skip_prk_steps', True),
            ('set_alpha_to_one', False),
            ('steps_offset', 1),
            ('_class_name', 'PNDMScheduler'),
            ('_diffusers_version', '0.8.0.dev0'),
            ('clip_sample', False)])
```

기존 스케줄러의 config를 호환 가능한 다른 스케줄러에 이식하는 것 역시 가능합니다.

다음 예시는 기존 스케줄러(`pipeline.scheduler`)를 다른 종류의 스케줄러(`DDIMScheduler`)로 바꾸는 코드입니다. 기존 스케줄러가 갖고 있던 config를 `.from_config` 메서드의 인자로 전달하는 것을 확인할 수 있습니다.

```python
from diffusers import DDIMScheduler

pipeline.scheduler = DDIMScheduler.from_config(pipeline.scheduler.config)
```



이제 파이프라인을 실행해서 두 스케줄러 사이의 생성된 이미지의 퀄리티를 비교해봅시다.

```python
generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator).images[0]
image
```

<p align="center">
    <br>
    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_ddim.png" width="400"/>
    <br>
</p>




## 스케줄러들 비교해보기

지금까지는 `PNDMScheduler`와 `DDIMScheduler` 스케줄러를 실행해보았습니다. 아직 비교해볼 스케줄러들이 더 많이 남아있으니 계속 비교해보도록 하겠습니다.



`LMSDiscreteScheduler`을 일반적으로 더 좋은 결과를 보여줍니다.

```python
from diffusers import LMSDiscreteScheduler

pipeline.scheduler = LMSDiscreteScheduler.from_config(pipeline.scheduler.config)

generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator).images[0]
image
```

<p align="center">
    <br>
    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_lms.png" width="400"/>
    <br>
</p>


`EulerDiscreteScheduler`와 `EulerAncestralDiscreteScheduler` 고작 30번의 inference step만으로도 높은 퀄리티의 이미지를 생성하는 것을 알 수 있습니다.

```python
from diffusers import EulerDiscreteScheduler

pipeline.scheduler = EulerDiscreteScheduler.from_config(pipeline.scheduler.config)

generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator, num_inference_steps=30).images[0]
image
```

<p align="center">
    <br>
    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_euler_discrete.png" width="400"/>
    <br>
</p>


```python
from diffusers import EulerAncestralDiscreteScheduler

pipeline.scheduler = EulerAncestralDiscreteScheduler.from_config(pipeline.scheduler.config)

generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator, num_inference_steps=30).images[0]
image
```

<p align="center">
    <br>
    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_euler_ancestral.png" width="400"/>
    <br>
</p>


지금 이 문서를 작성하는 현시점 기준에선, `DPMSolverMultistepScheduler`가 시간 대비 가장 좋은 품질의 이미지를 생성하는 것 같습니다. 20번 정도의 스텝만으로도 실행될 수 있습니다.



```python
from diffusers import DPMSolverMultistepScheduler

pipeline.scheduler = DPMSolverMultistepScheduler.from_config(pipeline.scheduler.config)

generator = torch.Generator(device="cuda").manual_seed(8)
image = pipeline(prompt, generator=generator, num_inference_steps=20).images[0]
image
```

<p align="center">
    <br>
    <img src="https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/diffusers_docs/astronaut_dpm.png" width="400"/>
    <br>
</p>


보시다시피 생성된 이미지들은 매우 비슷하고, 비슷한 퀄리티를 보이는 것 같습니다. 실제로 어떤 스케줄러를 선택할 것인가는 종종 특정 이용 사례에 기반해서 결정되곤 합니다. 결국 여러 종류의 스케줄러를 직접 실행시켜보고 눈으로 직접 비교해서 판단하는 게 좋은 선택일 것 같습니다.



## Flax에서 스케줄러 교체하기

JAX/Flax 사용자인 경우 기본 파이프라인 스케줄러를 변경할 수도 있습니다. 다음은 Flax Stable Diffusion 파이프라인과 초고속 [DDPM-Solver++ 스케줄러를](../api/schedulers/multistep_dpm_solver) 사용하여 추론을 실행하는 방법에 대한 예시입니다 .

```Python
import jax
import numpy as np
from flax.jax_utils import replicate
from flax.training.common_utils import shard

from diffusers import FlaxStableDiffusionPipeline, FlaxDPMSolverMultistepScheduler

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
scheduler, scheduler_state = FlaxDPMSolverMultistepScheduler.from_pretrained(
    model_id,
    subfolder="scheduler"
)
pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
    model_id,
    scheduler=scheduler,
    variant="bf16",
    dtype=jax.numpy.bfloat16,
)
params["scheduler"] = scheduler_state

# Generate 1 image per parallel device (8 on TPUv2-8 or TPUv3-8)
prompt = "a photo of an astronaut riding a horse on mars"
num_samples = jax.device_count()
prompt_ids = pipeline.prepare_inputs([prompt] * num_samples)

prng_seed = jax.random.PRNGKey(0)
num_inference_steps = 25

# shard inputs and rng
params = replicate(params)
prng_seed = jax.random.split(prng_seed, jax.device_count())
prompt_ids = shard(prompt_ids)

images = pipeline(prompt_ids, params, prng_seed, num_inference_steps, jit=True).images
images = pipeline.numpy_to_pil(np.asarray(images.reshape((num_samples,) + images.shape[-3:])))
```

> [!WARNING]
> 다음 Flax 스케줄러는 *아직* Flax Stable Diffusion 파이프라인과 호환되지 않습니다.
>
> - `FlaxLMSDiscreteScheduler`
> - `FlaxDDPMScheduler`



<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/schedulers.md" />

### 프롬프트에 가중치 부여하기
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/weighted_prompts.md

# 프롬프트에 가중치 부여하기


텍스트 가이드 기반의 diffusion 모델은 주어진 텍스트 프롬프트를 기반으로 이미지를 생성합니다.
텍스트 프롬프트에는 모델이 생성해야 하는 여러 개념이 포함될 수 있으며 프롬프트의 특정 부분에 가중치를 부여하는 것이 바람직한 경우가 많습니다.

Diffusion 모델은 문맥화된 텍스트 임베딩으로 diffusion 모델의 cross attention 레이어를 조절함으로써 작동합니다.
([더 많은 정보를 위한 Stable Diffusion Guide](https://huggingface.co/docs/optimum-neuron/main/en/package_reference/modeling#stable-diffusion)를 참고하세요).
따라서 프롬프트의 특정 부분을 강조하는(또는 강조하지 않는) 간단한 방법은 프롬프트의 관련 부분에 해당하는 텍스트 임베딩 벡터의 크기를 늘리거나 줄이는 것입니다.
이것은 "프롬프트 가중치 부여" 라고 하며, 커뮤니티에서 가장 요구하는 기능입니다.([이곳](https://github.com/huggingface/diffusers/issues/2431)의 issue를 보세요 ).

## Diffusers에서 프롬프트 가중치 부여하는 방법

우리는 `diffusers`의 역할이 다른 프로젝트를 가능하게 하는 필수적인 기능을 제공하는 toolbex라고 생각합니다.
[InvokeAI](https://github.com/invoke-ai/InvokeAI) 나 [diffuzers](https://github.com/abhishekkrthakur/diffuzers) 같은 강력한 UI를 구축할 수 있습니다.
프롬프트를 조작하는 방법을 지원하기 위해, `diffusers` 는
[StableDiffusionPipeline](https://huggingface.co/docs/diffusers/v0.18.2/en/api/pipelines/stable_diffusion/text2img#diffusers.StableDiffusionPipeline)와 같은
많은 파이프라인에 [prompt_embeds](https://huggingface.co/docs/diffusers/v0.14.0/en/api/pipelines/stable_diffusion/text2img#diffusers.StableDiffusionPipeline.__call__.prompt_embeds)
인수를 노출시켜, "prompt-weighted"/축척된 텍스트 임베딩을 파이프라인에 바로 전달할 수 있게 합니다.

[Compel 라이브러리](https://github.com/damian0815/compel)는 프롬프트의 일부를 강조하거나 강조하지 않을 수 있는 쉬운 방법을 제공합니다.
임베딩을 직접 준비하는 것 대신 이 방법을 사용하는 것을 강력히 추천합니다.

간단한 예제를 살펴보겠습니다.
다음과 같이 `"공을 갖고 노는 붉은색 고양이"` 이미지를 생성하고 싶습니다:

```py
from diffusers import StableDiffusionPipeline, UniPCMultistepScheduler

pipe = StableDiffusionPipeline.from_pretrained("CompVis/stable-diffusion-v1-4")
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)

prompt = "a red cat playing with a ball"

generator = torch.Generator(device="cpu").manual_seed(33)

image = pipe(prompt, generator=generator, num_inference_steps=20).images[0]
image
```

생성된 이미지:

![img](https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/compel/forest_0.png)

사진에서 알 수 있듯이, "공"은 이미지에 없습니다. 이 부분을 강조해 볼까요!

먼저 `compel` 라이브러리를 설치해야합니다:

```sh
pip install compel
```

그런 다음에는 `Compel` 오브젝트를 생성합니다:

```py
from compel import Compel

compel_proc = Compel(tokenizer=pipe.tokenizer, text_encoder=pipe.text_encoder)
```

이제 `"++"` 를 사용해서 "공" 을 강조해 봅시다:

```py
prompt = "a red cat playing with a ball++"
```

그리고 이 프롬프트를 파이프라인에 바로 전달하지 않고, `compel_proc` 를 사용하여 처리해야합니다:

```py
prompt_embeds = compel_proc(prompt)
```

파이프라인에 `prompt_embeds` 를 바로 전달할 수 있습니다:

```py
generator = torch.Generator(device="cpu").manual_seed(33)

images = pipe(prompt_embeds=prompt_embeds, generator=generator, num_inference_steps=20).images[0]
image
```

이제 "공"이 있는 그림을 출력할 수 있습니다!

![img](https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/compel/forest_1.png)

마찬가지로 `--` 접미사를 단어에 사용하여 문장의 일부를 강조하지 않을 수 있습니다. 한번 시도해 보세요!

즐겨찾는 파이프라인에 `prompt_embeds` 입력이 없는 경우 issue를 새로 만들어주세요.
Diffusers 팀은 최대한 대응하려고 노력합니다.

Compel 1.1.6 는 textual inversions을 사용하여 단순화하는 유티릴티 클래스를 추가합니다.
`DiffusersTextualInversionManager`를 인스턴스화 한 후 이를 Compel init에 전달합니다:

```
textual_inversion_manager = DiffusersTextualInversionManager(pipe)
compel = Compel(
    tokenizer=pipe.tokenizer,
    text_encoder=pipe.text_encoder,
    textual_inversion_manager=textual_inversion_manager)
```

더 많은 정보를 얻고 싶다면 [compel](https://github.com/damian0815/compel) 라이브러리 문서를 참고하세요.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/weighted_prompts.md" />

### Text-guided 이미지 인페인팅(inpainting)
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/inpaint.md

# Text-guided 이미지 인페인팅(inpainting)


`StableDiffusionInpaintPipeline`은 마스크와 텍스트 프롬프트를 제공하여 이미지의 특정 부분을 편집할 수 있도록 합니다. 이 기능은 인페인팅 작업을 위해 특별히 훈련된 [`runwayml/stable-diffusion-inpainting`](https://huggingface.co/runwayml/stable-diffusion-inpainting)과 같은 Stable Diffusion 버전을 사용합니다.

먼저 `StableDiffusionInpaintPipeline` 인스턴스를 불러옵니다:

```python
import PIL
import requests
import torch
from io import BytesIO

from diffusers import StableDiffusionInpaintPipeline

pipeline = StableDiffusionInpaintPipeline.from_pretrained(
    "runwayml/stable-diffusion-inpainting",
    torch_dtype=torch.float16,
)
pipeline = pipeline.to("cuda")
```

나중에 교체할 강아지 이미지와 마스크를 다운로드하세요:

```python
def download_image(url):
    response = requests.get(url)
    return PIL.Image.open(BytesIO(response.content)).convert("RGB")


img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"

init_image = download_image(img_url).resize((512, 512))
mask_image = download_image(mask_url).resize((512, 512))
```

이제 마스크를 다른 것으로 교체하라는 프롬프트를 만들 수 있습니다:

```python
prompt = "Face of a yellow cat, high resolution, sitting on a park bench"
image = pipe(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
```

`image`          | `mask_image` | `prompt` | output |
:-------------------------:|:-------------------------:|:-------------------------:|-------------------------:|
<img src="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png" alt="drawing" width="250"/> | <img src="https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png" alt="drawing" width="250"/> | ***Face of a yellow cat, high resolution, sitting on a park bench*** | <img src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/in_paint/yellow_cat_sitting_on_a_park_bench.png" alt="drawing" width="250"/> |

> [!WARNING]
> 이전의 실험적인 인페인팅 구현에서는 품질이 낮은 다른 프로세스를 사용했습니다. 이전 버전과의 호환성을 보장하기 위해 새 모델이 포함되지 않은 사전학습된 파이프라인을 불러오면 이전 인페인팅 방법이 계속 적용됩니다.

아래 Space에서 이미지 인페인팅을 직접 해보세요!

<iframe
  src="https://runwayml-stable-diffusion-inpainting.hf.space"
  frameborder="0"
  width="850"
  height="500"
></iframe>


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/inpaint.md" />

### Kandinsky
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/kandinsky.md

# Kandinsky


Kandinsky 모델은 일련의 다국어 text-to-image 생성 모델입니다. Kandinsky 2.0 모델은 두 개의 다국어 텍스트 인코더를 사용하고 그 결과를 연결해 UNet에 사용됩니다.

[Kandinsky 2.1](../api/pipelines/kandinsky)은 텍스트와 이미지 임베딩 간의 매핑을 생성하는 image prior 모델([`CLIP`](https://huggingface.co/docs/transformers/model_doc/clip))을 포함하도록 아키텍처를 변경했습니다. 이 매핑은 더 나은 text-image alignment를 제공하며, 학습 중에 텍스트 임베딩과 함께 사용되어 더 높은 품질의 결과를 가져옵니다. 마지막으로, Kandinsky 2.1은 spatial conditional 정규화 레이어를 추가하여 사실감을 높여주는 [Modulating Quantized Vectors (MoVQ)](https://huggingface.co/papers/2209.09002) 디코더를 사용하여 latents를 이미지로 디코딩합니다.

[Kandinsky 2.2](../api/pipelines/kandinsky_v22)는 image prior 모델의 이미지 인코더를 더 큰 CLIP-ViT-G 모델로 교체하여 품질을 개선함으로써 이전 모델을 개선했습니다. 또한 image prior 모델은 해상도와 종횡비가 다른 이미지로 재훈련되어 더 높은 해상도의 이미지와 다양한 이미지 크기를 생성합니다.

[Kandinsky 3](../api/pipelines/kandinsky3)는 아키텍처를 단순화하고 prior 모델과 diffusion 모델을 포함하는 2단계 생성 프로세스에서 벗어나고 있습니다. 대신, Kandinsky 3는 [Flan-UL2](https://huggingface.co/google/flan-ul2)를 사용하여 텍스트를 인코딩하고, [BigGan-deep](https://hf.co/papers/1809.11096) 블록이 포함된 UNet을 사용하며, [Sber-MoVQGAN](https://github.com/ai-forever/MoVQGAN)을 사용하여 latents를 이미지로 디코딩합니다. 텍스트 이해와 생성된 이미지 품질은 주로 더 큰 텍스트 인코더와 UNet을 사용함으로써 달성됩니다.

이 가이드에서는 text-to-image, image-to-image, 인페인팅, 보간 등을 위해 Kandinsky 모델을 사용하는 방법을 설명합니다.

시작하기 전에 다음 라이브러리가 설치되어 있는지 확인하세요:

```py
# Colab에서 필요한 라이브러리를 설치하기 위해 주석을 제외하세요
#!pip install -q diffusers transformers accelerate
```

> [!WARNING]
> Kandinsky 2.1과 2.2의 사용법은 매우 유사합니다! 유일한 차이점은 Kandinsky 2.2는 latents를 디코딩할 때 `프롬프트`를 입력으로 받지 않는다는 것입니다. 대신, Kandinsky 2.2는 디코딩 중에는 `image_embeds`만 받아들입니다.
>
> <br>
>
> Kandinsky 3는 더 간결한 아키텍처를 가지고 있으며 prior 모델이 필요하지 않습니다. 즉, [Stable Diffusion XL](sdxl)과 같은 다른 diffusion 모델과 사용법이 동일합니다.

## Text-to-image

모든 작업에 Kandinsky 모델을 사용하려면 항상 프롬프트를 인코딩하고 이미지 임베딩을 생성하는 prior 파이프라인을 설정하는 것부터 시작해야 합니다. 이전 파이프라인은 negative 프롬프트 `""`에 해당하는 `negative_image_embeds`도 생성합니다. 더 나은 결과를 얻으려면 이전 파이프라인에 실제 `negative_prompt`를 전달할 수 있지만, 이렇게 하면 prior 파이프라인의 유효 배치 크기가 2배로 증가합니다.

<hfoptions id="text-to-image">
<hfoption id="Kandinsky 2.1">

```py
from diffusers import KandinskyPriorPipeline, KandinskyPipeline
import torch

prior_pipeline = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16).to("cuda")
pipeline = KandinskyPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16).to("cuda")

prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
negative_prompt = "low quality, bad quality" # negative 프롬프트 포함은 선택적이지만, 보통 결과는 더 좋습니다
image_embeds, negative_image_embeds = prior_pipeline(prompt, negative_prompt, guidance_scale=1.0).to_tuple()
```

이제 모든 프롬프트와 임베딩을 `KandinskyPipeline`에 전달하여 이미지를 생성합니다:

```py
image = pipeline(prompt, image_embeds=image_embeds, negative_prompt=negative_prompt, negative_image_embeds=negative_image_embeds, height=768, width=768).images[0]
image
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/cheeseburger.png"/>
</div>

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
from diffusers import KandinskyV22PriorPipeline, KandinskyV22Pipeline
import torch

prior_pipeline = KandinskyV22PriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float16).to("cuda")
pipeline = KandinskyV22Pipeline.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16).to("cuda")

prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
negative_prompt = "low quality, bad quality" # negative 프롬프트 포함은 선택적이지만, 보통 결과는 더 좋습니다
image_embeds, negative_image_embeds = prior_pipeline(prompt, guidance_scale=1.0).to_tuple()
```

이미지 생성을 위해 `image_embeds`와 `negative_image_embeds`를 `KandinskyV22Pipeline`에 전달합니다:

```py
image = pipeline(image_embeds=image_embeds, negative_image_embeds=negative_image_embeds, height=768, width=768).images[0]
image
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-text-to-image.png"/>
</div>

</hfoption>
<hfoption id="Kandinsky 3">

Kandinsky 3는 prior 모델이 필요하지 않으므로 `Kandinsky3Pipeline`을 직접 불러오고 이미지 생성 프롬프트를 전달할 수 있습니다:

```py
from diffusers import Kandinsky3Pipeline
import torch

pipeline = Kandinsky3Pipeline.from_pretrained("kandinsky-community/kandinsky-3", variant="fp16", torch_dtype=torch.float16)
pipeline.enable_model_cpu_offload()

prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
image = pipeline(prompt).images[0]
image
```

</hfoption>
</hfoptions>

🤗 Diffusers는 또한 `KandinskyCombinedPipeline` 및 `KandinskyV22CombinedPipeline`이 포함된 end-to-end API를 제공하므로 prior 파이프라인과 text-to-image 변환 파이프라인을 별도로 불러올 필요가 없습니다. 결합된 파이프라인은 prior 모델과 디코더를 모두 자동으로 불러옵니다. 원하는 경우 `prior_guidance_scale` 및 `prior_num_inference_steps` 매개 변수를 사용하여 prior 파이프라인에 대해 다른 값을 설정할 수 있습니다.

내부에서 결합된 파이프라인을 자동으로 호출하려면 `AutoPipelineForText2Image`를 사용합니다:

<hfoptions id="text-to-image">
<hfoption id="Kandinsky 2.1">

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
pipeline.enable_model_cpu_offload()

prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
negative_prompt = "low quality, bad quality"

image = pipeline(prompt=prompt, negative_prompt=negative_prompt, prior_guidance_scale=1.0, guidance_scale=4.0, height=768, width=768).images[0]
image
```

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16)
pipeline.enable_model_cpu_offload()

prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
negative_prompt = "low quality, bad quality"

image = pipeline(prompt=prompt, negative_prompt=negative_prompt, prior_guidance_scale=1.0, guidance_scale=4.0, height=768, width=768).images[0]
image
```

</hfoption>
</hfoptions>

## Image-to-image

Image-to-image 경우, 초기 이미지와 텍스트 프롬프트를 전달하여 파이프라인에 이미지를 conditioning합니다. Prior 파이프라인을 불러오는 것으로 시작합니다:

<hfoptions id="image-to-image">
<hfoption id="Kandinsky 2.1">

```py
import torch
from diffusers import KandinskyImg2ImgPipeline, KandinskyPriorPipeline

prior_pipeline = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
pipeline = KandinskyImg2ImgPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
```

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
import torch
from diffusers import KandinskyV22Img2ImgPipeline, KandinskyPriorPipeline

prior_pipeline = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
pipeline = KandinskyV22Img2ImgPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
```

</hfoption>
<hfoption id="Kandinsky 3">

Kandinsky 3는 prior 모델이 필요하지 않으므로 image-to-image 파이프라인을 직접 불러올 수 있습니다:

```py
from diffusers import Kandinsky3Img2ImgPipeline
from diffusers.utils import load_image
import torch

pipeline = Kandinsky3Img2ImgPipeline.from_pretrained("kandinsky-community/kandinsky-3", variant="fp16", torch_dtype=torch.float16)
pipeline.enable_model_cpu_offload()
```

</hfoption>
</hfoptions>

Conditioning할 이미지를 다운로드합니다:

```py
from diffusers.utils import load_image

# 이미지 다운로드
url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
original_image = load_image(url)
original_image = original_image.resize((768, 512))
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"/>
</div>

Prior 파이프라인으로 `image_embeds`와 `negative_image_embeds`를 생성합니다:

```py
prompt = "A fantasy landscape, Cinematic lighting"
negative_prompt = "low quality, bad quality"

image_embeds, negative_image_embeds = prior_pipeline(prompt, negative_prompt).to_tuple()
```

이제 원본 이미지와 모든 프롬프트 및 임베딩을 파이프라인으로 전달하여 이미지를 생성합니다:

<hfoptions id="image-to-image">
<hfoption id="Kandinsky 2.1">

```py
from diffusers.utils import make_image_grid

image = pipeline(prompt, negative_prompt=negative_prompt, image=original_image, image_embeds=image_embeds, negative_image_embeds=negative_image_embeds, height=768, width=768, strength=0.3).images[0]
make_image_grid([original_image.resize((512, 512)), image.resize((512, 512))], rows=1, cols=2)
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/img2img_fantasyland.png"/>
</div>

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
from diffusers.utils import make_image_grid

image = pipeline(image=original_image, image_embeds=image_embeds, negative_image_embeds=negative_image_embeds, height=768, width=768, strength=0.3).images[0]
make_image_grid([original_image.resize((512, 512)), image.resize((512, 512))], rows=1, cols=2)
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-image-to-image.png"/>
</div>

</hfoption>
<hfoption id="Kandinsky 3">

```py
image = pipeline(prompt, negative_prompt=negative_prompt, image=image, strength=0.75, num_inference_steps=25).images[0]
image
```

</hfoption>
</hfoptions>

또한 🤗 Diffusers에서는 `KandinskyImg2ImgCombinedPipeline` 및 `KandinskyV22Img2ImgCombinedPipeline`이 포함된 end-to-end API를 제공하므로 prior 파이프라인과 image-to-image 파이프라인을 별도로 불러올 필요가 없습니다. 결합된 파이프라인은 prior 모델과 디코더를 모두 자동으로 불러옵니다. 원하는 경우 `prior_guidance_scale` 및 `prior_num_inference_steps` 매개 변수를 사용하여 이전 파이프라인에 대해 다른 값을 설정할 수 있습니다.

내부에서 결합된 파이프라인을 자동으로 호출하려면 `AutoPipelineForImage2Image`를 사용합니다:

<hfoptions id="image-to-image">
<hfoption id="Kandinsky 2.1">

```py
from diffusers import AutoPipelineForImage2Image
from diffusers.utils import make_image_grid, load_image
import torch

pipeline = AutoPipelineForImage2Image.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16, use_safetensors=True)
pipeline.enable_model_cpu_offload()

prompt = "A fantasy landscape, Cinematic lighting"
negative_prompt = "low quality, bad quality"

url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
original_image = load_image(url)

original_image.thumbnail((768, 768))

image = pipeline(prompt=prompt, negative_prompt=negative_prompt, image=original_image, strength=0.3).images[0]
make_image_grid([original_image.resize((512, 512)), image.resize((512, 512))], rows=1, cols=2)
```

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
from diffusers import AutoPipelineForImage2Image
from diffusers.utils import make_image_grid, load_image
import torch

pipeline = AutoPipelineForImage2Image.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16)
pipeline.enable_model_cpu_offload()

prompt = "A fantasy landscape, Cinematic lighting"
negative_prompt = "low quality, bad quality"

url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
original_image = load_image(url)

original_image.thumbnail((768, 768))

image = pipeline(prompt=prompt, negative_prompt=negative_prompt, image=original_image, strength=0.3).images[0]
make_image_grid([original_image.resize((512, 512)), image.resize((512, 512))], rows=1, cols=2)
```

</hfoption>
</hfoptions>

## Inpainting

> [!WARNING]
> ⚠️ Kandinsky 모델은 이제 검은색 픽셀 대신 ⬜️ **흰색 픽셀**을 사용하여 마스크 영역을 표현합니다. 프로덕션에서 `KandinskyInpaintPipeline`을 사용하는 경우 흰색 픽셀을 사용하도록 마스크를 변경해야 합니다:
>
> ```py
> # PIL 입력에 대해
> import PIL.ImageOps
> mask = PIL.ImageOps.invert(mask)
>
> # PyTorch와 NumPy 입력에 대해
> mask = 1 - mask
> ```

인페인팅에서는 원본 이미지, 원본 이미지에서 대체할 영역의 마스크, 인페인팅할 내용에 대한 텍스트 프롬프트가 필요합니다. Prior 파이프라인을 불러옵니다:

<hfoptions id="inpaint">
<hfoption id="Kandinsky 2.1">

```py
from diffusers import KandinskyInpaintPipeline, KandinskyPriorPipeline
from diffusers.utils import load_image, make_image_grid
import torch
import numpy as np
from PIL import Image

prior_pipeline = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
pipeline = KandinskyInpaintPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-inpaint", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
```

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
from diffusers import KandinskyV22InpaintPipeline, KandinskyV22PriorPipeline
from diffusers.utils import load_image, make_image_grid
import torch
import numpy as np
from PIL import Image

prior_pipeline = KandinskyV22PriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
pipeline = KandinskyV22InpaintPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-decoder-inpaint", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
```

</hfoption>
</hfoptions>

초기 이미지를 불러오고 마스크를 생성합니다:

```py
init_image = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png")
mask = np.zeros((768, 768), dtype=np.float32)
# mask area above cat's head
mask[:250, 250:-250] = 1
```

Prior 파이프라인으로 임베딩을 생성합니다:

```py
prompt = "a hat"
prior_output = prior_pipeline(prompt)
```

이제 이미지 생성을 위해 초기 이미지, 마스크, 프롬프트와 임베딩을 파이프라인에 전달합니다:

<hfoptions id="inpaint">
<hfoption id="Kandinsky 2.1">

```py
output_image = pipeline(prompt, image=init_image, mask_image=mask, **prior_output, height=768, width=768, num_inference_steps=150).images[0]
mask = Image.fromarray((mask*255).astype('uint8'), 'L')
make_image_grid([init_image, mask, output_image], rows=1, cols=3)
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/inpaint_cat_hat.png"/>
</div>

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
output_image = pipeline(image=init_image, mask_image=mask, **prior_output, height=768, width=768, num_inference_steps=150).images[0]
mask = Image.fromarray((mask*255).astype('uint8'), 'L')
make_image_grid([init_image, mask, output_image], rows=1, cols=3)
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinskyv22-inpaint.png"/>
</div>

</hfoption>
</hfoptions>

`KandinskyInpaintCombinedPipeline` 및 `KandinskyV22InpaintCombinedPipeline`을 사용하여 내부에서 prior 및 디코더 파이프라인을 함께 호출할 수 있습니다. 이를 위해 `AutoPipelineForInpainting`을 사용합니다:

<hfoptions id="inpaint">
<hfoption id="Kandinsky 2.1">

```py
import torch
import numpy as np
from PIL import Image
from diffusers import AutoPipelineForInpainting
from diffusers.utils import load_image, make_image_grid

pipe = AutoPipelineForInpainting.from_pretrained("kandinsky-community/kandinsky-2-1-inpaint", torch_dtype=torch.float16)
pipe.enable_model_cpu_offload()

init_image = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png")
mask = np.zeros((768, 768), dtype=np.float32)
# 고양이 머리 위 마스크 지역
mask[:250, 250:-250] = 1
prompt = "a hat"

output_image = pipe(prompt=prompt, image=init_image, mask_image=mask).images[0]
mask = Image.fromarray((mask*255).astype('uint8'), 'L')
make_image_grid([init_image, mask, output_image], rows=1, cols=3)
```

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
import torch
import numpy as np
from PIL import Image
from diffusers import AutoPipelineForInpainting
from diffusers.utils import load_image, make_image_grid

pipe = AutoPipelineForInpainting.from_pretrained("kandinsky-community/kandinsky-2-2-decoder-inpaint", torch_dtype=torch.float16)
pipe.enable_model_cpu_offload()

init_image = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png")
mask = np.zeros((768, 768), dtype=np.float32)
# 고양이 머리 위 마스크 영역
mask[:250, 250:-250] = 1
prompt = "a hat"

output_image = pipe(prompt=prompt, image=original_image, mask_image=mask).images[0]
mask = Image.fromarray((mask*255).astype('uint8'), 'L')
make_image_grid([init_image, mask, output_image], rows=1, cols=3)
```

</hfoption>
</hfoptions>

## Interpolation (보간)

Interpolation(보간)을 사용하면 이미지와 텍스트 임베딩 사이의 latent space를 탐색할 수 있어 prior 모델의 중간 결과물을 볼 수 있는 멋진 방법입니다. Prior 파이프라인과 보간하려는 두 개의 이미지를 불러옵니다:

<hfoptions id="interpolate">
<hfoption id="Kandinsky 2.1">

```py
from diffusers import KandinskyPriorPipeline, KandinskyPipeline
from diffusers.utils import load_image, make_image_grid
import torch

prior_pipeline = KandinskyPriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
img_1 = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png")
img_2 = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/starry_night.jpeg")
make_image_grid([img_1.resize((512,512)), img_2.resize((512,512))], rows=1, cols=2)
```

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
from diffusers import KandinskyV22PriorPipeline, KandinskyV22Pipeline
from diffusers.utils import load_image, make_image_grid
import torch

prior_pipeline = KandinskyV22PriorPipeline.from_pretrained("kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
img_1 = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png")
img_2 = load_image("https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/starry_night.jpeg")
make_image_grid([img_1.resize((512,512)), img_2.resize((512,512))], rows=1, cols=2)
```

</hfoption>
</hfoptions>

<div class="flex gap-4">
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">a cat</figcaption>
  </div>
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/starry_night.jpeg"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">Van Gogh's Starry Night painting</figcaption>
  </div>
</div>

보간할 텍스트 또는 이미지를 지정하고 각 텍스트 또는 이미지에 대한 가중치를 설정합니다. 가중치를 실험하여 보간에 어떤 영향을 미치는지 확인하세요!

```py
images_texts = ["a cat", img_1, img_2]
weights = [0.3, 0.3, 0.4]
```

`interpolate` 함수를 호출하여 임베딩을 생성한 다음, 파이프라인으로 전달하여 이미지를 생성합니다:

<hfoptions id="interpolate">
<hfoption id="Kandinsky 2.1">

```py
# 프롬프트는 빈칸으로 남겨도 됩니다
prompt = ""
prior_out = prior_pipeline.interpolate(images_texts, weights)

pipeline = KandinskyPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16, use_safetensors=True).to("cuda")

image = pipeline(prompt, **prior_out, height=768, width=768).images[0]
image
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinsky-docs/starry_cat.png"/>
</div>

</hfoption>
<hfoption id="Kandinsky 2.2">

```py
# 프롬프트는 빈칸으로 남겨도 됩니다
prompt = ""
prior_out = prior_pipeline.interpolate(images_texts, weights)

pipeline = KandinskyV22Pipeline.from_pretrained("kandinsky-community/kandinsky-2-2-decoder", torch_dtype=torch.float16, use_safetensors=True).to("cuda")

image = pipeline(prompt, **prior_out, height=768, width=768).images[0]
image
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/kandinskyv22-interpolate.png"/>
</div>

</hfoption>
</hfoptions>

## ControlNet

> [!WARNING]
> ⚠️ ControlNet은 Kandinsky 2.2에서만 지원됩니다!

ControlNet을 사용하면 depth map이나 edge detection와 같은 추가 입력을 통해 사전학습된 large diffusion 모델을 conditioning할 수 있습니다. 예를 들어, 모델이 depth map의 구조를 이해하고 보존할 수 있도록 깊이 맵으로 Kandinsky 2.2를 conditioning할 수 있습니다.

이미지를 불러오고 depth map을 추출해 보겠습니다:

```py
from diffusers.utils import load_image

img = load_image(
    "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinskyv22/cat.png"
).resize((768, 768))
img
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinskyv22/cat.png"/>
</div>

그런 다음 🤗 Transformers의 `depth-estimation` `Pipeline`을 사용하여 이미지를 처리해 depth map을 구할 수 있습니다:

```py
import torch
import numpy as np

from transformers import pipeline

def make_hint(image, depth_estimator):
    image = depth_estimator(image)["depth"]
    image = np.array(image)
    image = image[:, :, None]
    image = np.concatenate([image, image, image], axis=2)
    detected_map = torch.from_numpy(image).float() / 255.0
    hint = detected_map.permute(2, 0, 1)
    return hint

depth_estimator = pipeline("depth-estimation")
hint = make_hint(img, depth_estimator).unsqueeze(0).half().to("cuda")
```

### Text-to-image [[controlnet-text-to-image]]

Prior 파이프라인과 `KandinskyV22ControlnetPipeline`를 불러옵니다:

```py
from diffusers import KandinskyV22PriorPipeline, KandinskyV22ControlnetPipeline

prior_pipeline = KandinskyV22PriorPipeline.from_pretrained(
    "kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float16, use_safetensors=True
).to("cuda")

pipeline = KandinskyV22ControlnetPipeline.from_pretrained(
    "kandinsky-community/kandinsky-2-2-controlnet-depth", torch_dtype=torch.float16
).to("cuda")
```

프롬프트와 negative 프롬프트로 이미지 임베딩을 생성합니다:

```py
prompt = "A robot, 4k photo"
negative_prior_prompt = "lowres, text, error, cropped, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, username, watermark, signature"

generator = torch.Generator(device="cuda").manual_seed(43)

image_emb, zero_image_emb = prior_pipeline(
    prompt=prompt, negative_prompt=negative_prior_prompt, generator=generator
).to_tuple()
```

마지막으로 이미지 임베딩과 depth 이미지를 `KandinskyV22ControlnetPipeline`에 전달하여 이미지를 생성합니다:

```py
image = pipeline(image_embeds=image_emb, negative_image_embeds=zero_image_emb, hint=hint, num_inference_steps=50, generator=generator, height=768, width=768).images[0]
image
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinskyv22/robot_cat_text2img.png"/>
</div>

### Image-to-image [[controlnet-image-to-image]]

ControlNet을 사용한 image-to-image의 경우, 다음을 사용할 필요가 있습니다:

- `KandinskyV22PriorEmb2EmbPipeline`로 텍스트 프롬프트와 이미지에서 이미지 임베딩을 생성합니다.
- `KandinskyV22ControlnetImg2ImgPipeline`로 초기 이미지와 이미지 임베딩에서 이미지를 생성합니다.

🤗 Transformers에서 `depth-estimation` `Pipeline`을 사용하여 고양이의 초기 이미지의 depth map을 처리해 추출합니다:

```py
import torch
import numpy as np

from diffusers import KandinskyV22PriorEmb2EmbPipeline, KandinskyV22ControlnetImg2ImgPipeline
from diffusers.utils import load_image
from transformers import pipeline

img = load_image(
    "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinskyv22/cat.png"
).resize((768, 768))

def make_hint(image, depth_estimator):
    image = depth_estimator(image)["depth"]
    image = np.array(image)
    image = image[:, :, None]
    image = np.concatenate([image, image, image], axis=2)
    detected_map = torch.from_numpy(image).float() / 255.0
    hint = detected_map.permute(2, 0, 1)
    return hint

depth_estimator = pipeline("depth-estimation")
hint = make_hint(img, depth_estimator).unsqueeze(0).half().to("cuda")
```

Prior 파이프라인과 `KandinskyV22ControlnetImg2ImgPipeline`을 불러옵니다:

```py
prior_pipeline = KandinskyV22PriorEmb2EmbPipeline.from_pretrained(
    "kandinsky-community/kandinsky-2-2-prior", torch_dtype=torch.float16, use_safetensors=True
).to("cuda")

pipeline = KandinskyV22ControlnetImg2ImgPipeline.from_pretrained(
    "kandinsky-community/kandinsky-2-2-controlnet-depth", torch_dtype=torch.float16
).to("cuda")
```

텍스트 프롬프트와 초기 이미지를 이전 파이프라인에 전달하여 이미지 임베딩을 생성합니다:

```py
prompt = "A robot, 4k photo"
negative_prior_prompt = "lowres, text, error, cropped, worst quality, low quality, jpeg artifacts, ugly, duplicate, morbid, mutilated, out of frame, extra fingers, mutated hands, poorly drawn hands, poorly drawn face, mutation, deformed, blurry, dehydrated, bad anatomy, bad proportions, extra limbs, cloned face, disfigured, gross proportions, malformed limbs, missing arms, missing legs, extra arms, extra legs, fused fingers, too many fingers, long neck, username, watermark, signature"

generator = torch.Generator(device="cuda").manual_seed(43)

img_emb = prior_pipeline(prompt=prompt, image=img, strength=0.85, generator=generator)
negative_emb = prior_pipeline(prompt=negative_prior_prompt, image=img, strength=1, generator=generator)
```

이제 `KandinskyV22ControlnetImg2ImgPipeline`을 실행하여 초기 이미지와 이미지 임베딩으로부터 이미지를 생성할 수 있습니다:

```py
image = pipeline(image=img, strength=0.5, image_embeds=img_emb.image_embeds, negative_image_embeds=negative_emb.image_embeds, hint=hint, num_inference_steps=50, generator=generator, height=768, width=768).images[0]
make_image_grid([img.resize((512, 512)), image.resize((512, 512))], rows=1, cols=2)
```

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinskyv22/robot_cat.png"/>
</div>

## 최적화

Kandinsky는 mapping을 생성하기 위한 prior 파이프라인과 latents를 이미지로 디코딩하기 위한 두 번째 파이프라인이 필요하다는 점에서 독특합니다. 대부분의 계산이 두 번째 파이프라인에서 이루어지므로 최적화의 노력은 두 번째 파이프라인에 집중되어야 합니다. 다음은 추론 중 Kandinsky키를 개선하기 위한 몇 가지 팁입니다.

1. PyTorch < 2.0을 사용할 경우 [xFormers](../optimization/xformers)을 활성화합니다.

```diff
  from diffusers import DiffusionPipeline
  import torch

  pipe = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
+ pipe.enable_xformers_memory_efficient_attention()
```

2. PyTorch >= 2.0을 사용할 경우 `torch.compile`을 활성화하여 scaled dot-product attention (SDPA)를 자동으로 사용하도록 합니다:

```diff
  pipe.unet.to(memory_format=torch.channels_last)
+ pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
```

이는 attention processor를 명시적으로 `AttnAddedKVProcessor2_0`을 사용하도록 설정하는 것과 동일합니다:

```py
from diffusers.models.attention_processor import AttnAddedKVProcessor2_0

pipe.unet.set_attn_processor(AttnAddedKVProcessor2_0())
```

3. 메모리 부족 오류를 방지하기 위해 `enable_model_cpu_offload()`를 사용하여 모델을 CPU로 오프로드합니다:

```diff
  from diffusers import DiffusionPipeline
  import torch

  pipe = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
+ pipe.enable_model_cpu_offload()
```

4. 기본적으로 text-to-image 파이프라인은 `DDIMScheduler`를 사용하지만, `DDPMScheduler`와 같은 다른 스케줄러로 대체하여 추론 속도와 이미지 품질 간의 균형에 어떤 영향을 미치는지 확인할 수 있습니다:

```py
from diffusers import DDPMScheduler
from diffusers import DiffusionPipeline

scheduler = DDPMScheduler.from_pretrained("kandinsky-community/kandinsky-2-1", subfolder="ddpm_scheduler")
pipe = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", scheduler=scheduler, torch_dtype=torch.float16, use_safetensors=True).to("cuda")
```


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/kandinsky.md" />

### 제어된 생성
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/controlling_generation.md

# 제어된 생성

Diffusion 모델에 의해 생성된 출력을 제어하는 것은 커뮤니티에서 오랫동안 추구해 왔으며 현재 활발한 연구 주제입니다. 널리 사용되는 많은 diffusion 모델에서는 이미지와 텍스트 프롬프트 등 입력의 미묘한 변화로 인해 출력이 크게 달라질 수 있습니다. 이상적인 세계에서는 의미가 유지되고 변경되는 방식을 제어할 수 있기를 원합니다.

의미 보존의 대부분의 예는 입력의 변화를 출력의 변화에 정확하게 매핑하는 것으로 축소됩니다. 즉, 프롬프트에서 피사체에 형용사를 추가하면 전체 이미지가 보존되고 변경된 피사체만 수정됩니다. 또는 특정 피사체의 이미지를 변형하면 피사체의 포즈가 유지됩니다.

추가적으로 생성된 이미지의 품질에는 의미 보존 외에도 영향을 미치고자 하는 품질이 있습니다. 즉, 일반적으로 결과물의 품질이 좋거나 특정 스타일을 고수하거나 사실적이기를 원합니다.

diffusion 모델 생성을 제어하기 위해 `diffusers`가 지원하는 몇 가지 기술을 문서화합니다. 많은 부분이 최첨단 연구이며 미묘한 차이가 있을 수 있습니다. 명확한 설명이 필요하거나 제안 사항이 있으면 주저하지 마시고 [포럼](https://discuss.huggingface.co/) 또는 [GitHub 이슈](https://github.com/huggingface/diffusers/issues)에서 토론을 시작하세요.

생성 제어 방법에 대한 개략적인 설명과 기술 개요를 제공합니다. 기술에 대한 자세한 설명은 파이프라인에서 링크된 원본 논문을 참조하는 것이 가장 좋습니다.

사용 사례에 따라 적절한 기술을 선택해야 합니다. 많은 경우 이러한 기법을 결합할 수 있습니다. 예를 들어, 텍스트 반전과 SEGA를 결합하여 텍스트 반전을 사용하여 생성된 출력에 더 많은 의미적 지침을 제공할 수 있습니다.

별도의 언급이 없는 한, 이러한 기법은 기존 모델과 함께 작동하며 자체 가중치가 필요하지 않은 기법입니다.

1. [Instruct Pix2Pix](#instruct-pix2pix)
2. [Pix2Pix Zero](#pix2pixzero)
3. [Attend and Excite](#attend-and-excite)
4. [Semantic Guidance](#semantic-guidance)
5. [Self-attention Guidance](#self-attention-guidance)
6. [Depth2Image](#depth2image)
7. [MultiDiffusion Panorama](#multidiffusion-panorama)
8. [DreamBooth](#dreambooth)
9. [Textual Inversion](#textual-inversion)
10. [ControlNet](#controlnet)
11. [Prompt Weighting](#prompt-weighting)
12. [Custom Diffusion](#custom-diffusion)
13. [Model Editing](#model-editing)
14. [DiffEdit](#diffedit)
15. [T2I-Adapter](#t2i-adapter)

편의를 위해, 추론만 하거나 파인튜닝/학습하는 방법에 대한 표를 제공합니다.

|                     **Method**                      | **Inference only** | **Requires training /<br> fine-tuning** |                                          **Comments**                                           |
| :-------------------------------------------------: | :----------------: | :-------------------------------------: | :---------------------------------------------------------------------------------------------: |
|        [Instruct Pix2Pix](#instruct-pix2pix)        |         ✅         |                   ❌                    | Can additionally be<br>fine-tuned for better <br>performance on specific <br>edit instructions. |
|            [Pix2Pix Zero](#pix2pixzero)             |         ✅         |                   ❌                    |                                                                                                 |
|       [Attend and Excite](#attend-and-excite)       |         ✅         |                   ❌                    |                                                                                                 |
|       [Semantic Guidance](#semantic-guidance)       |         ✅         |                   ❌                    |                                                                                                 |
| [Self-attention Guidance](#self-attention-guidance) |         ✅         |                   ❌                    |                                                                                                 |
|             [Depth2Image](#depth2image)             |         ✅         |                   ❌                    |                                                                                                 |
| [MultiDiffusion Panorama](#multidiffusion-panorama) |         ✅         |                   ❌                    |                                                                                                 |
|              [DreamBooth](#dreambooth)              |         ❌         |                   ✅                    |                                                                                                 |
|       [Textual Inversion](#textual-inversion)       |         ❌         |                   ✅                    |                                                                                                 |
|              [ControlNet](#controlnet)              |         ✅         |                   ❌                    |             A ControlNet can be <br>trained/fine-tuned on<br>a custom conditioning.             |
|        [Prompt Weighting](#prompt-weighting)        |         ✅         |                   ❌                    |                                                                                                 |
|        [Custom Diffusion](#custom-diffusion)        |         ❌         |                   ✅                    |                                                                                                 |
|           [Model Editing](#model-editing)           |         ✅         |                   ❌                    |                                                                                                 |
|                [DiffEdit](#diffedit)                |         ✅         |                   ❌                    |                                                                                                 |
|             [T2I-Adapter](#t2i-adapter)             |         ✅         |                   ❌                    |                                                                                                 |

## Pix2Pix Instruct

[Paper](https://huggingface.co/papers/2211.09800)

[Instruct Pix2Pix](../api/pipelines/stable_diffusion/pix2pix) 는 입력 이미지 편집을 지원하기 위해 stable diffusion에서 미세-조정되었습니다. 이미지와 편집을 설명하는 프롬프트를 입력으로 받아 편집된 이미지를 출력합니다.
Instruct Pix2Pix는 [InstructGPT](https://openai.com/blog/instruction-following/)와 같은 프롬프트와 잘 작동하도록 명시적으로 훈련되었습니다.

사용 방법에 대한 자세한 내용은 [여기](../api/pipelines/stable_diffusion/pix2pix)를 참조하세요.

## Pix2Pix Zero

[Paper](https://huggingface.co/papers/2302.03027)

[Pix2Pix Zero](../api/pipelines/stable_diffusion/pix2pix_zero)를 사용하면 일반적인 이미지 의미를 유지하면서 한 개념이나 피사체가 다른 개념이나 피사체로 변환되도록 이미지를 수정할 수 있습니다.

노이즈 제거 프로세스는 한 개념적 임베딩에서 다른 개념적 임베딩으로 안내됩니다. 중간 잠복(intermediate latents)은 디노이징(denoising?) 프로세스 중에 최적화되어 참조 주의 지도(reference attention maps)를 향해 나아갑니다. 참조 주의 지도(reference attention maps)는 입력 이미지의 노이즈 제거(?) 프로세스에서 나온 것으로 의미 보존을 장려하는 데 사용됩니다.

Pix2Pix Zero는 합성 이미지와 실제 이미지를 편집하는 데 모두 사용할 수 있습니다.

- 합성 이미지를 편집하려면 먼저 캡션이 지정된 이미지를 생성합니다.
  다음으로 편집할 컨셉과 새로운 타겟 컨셉에 대한 이미지 캡션을 생성합니다. 이를 위해 [Flan-T5](https://huggingface.co/docs/transformers/model_doc/flan-t5)와 같은 모델을 사용할 수 있습니다. 그런 다음 텍스트 인코더를 통해 소스 개념과 대상 개념 모두에 대한 "평균" 프롬프트 임베딩을 생성합니다. 마지막으로, 합성 이미지를 편집하기 위해 pix2pix-zero 알고리즘을 사용합니다.
- 실제 이미지를 편집하려면 먼저 [BLIP](https://huggingface.co/docs/transformers/model_doc/blip)과 같은 모델을 사용하여 이미지 캡션을 생성합니다. 그런 다음 프롬프트와 이미지에 ddim 반전을 적용하여 "역(inverse)" latents을 생성합니다. 이전과 마찬가지로 소스 및 대상 개념 모두에 대한 "평균(mean)" 프롬프트 임베딩이 생성되고 마지막으로 "역(inverse)" latents와 결합된 pix2pix-zero 알고리즘이 이미지를 편집하는 데 사용됩니다.

> [!TIP]
> Pix2Pix Zero는 '제로 샷(zero-shot)' 이미지 편집이 가능한 최초의 모델입니다.
> 즉, 이 모델은 다음과 같이 일반 소비자용 GPU에서 1분 이내에 이미지를 편집할 수 있습니다(../api/pipelines/stable_diffusion/pix2pix_zero#usage-example).

위에서 언급했듯이 Pix2Pix Zero에는 특정 개념으로 세대를 유도하기 위해 (UNet, VAE 또는 텍스트 인코더가 아닌) latents을 최적화하는 기능이 포함되어 있습니다.즉, 전체 파이프라인에 표준 [StableDiffusionPipeline](../api/pipelines/stable_diffusion/text2img)보다 더 많은 메모리가 필요할 수 있습니다.

사용 방법에 대한 자세한 내용은 [여기](../api/pipelines/stable_diffusion/pix2pix_zero)를 참조하세요.

## Attend and Excite

[Paper](https://huggingface.co/papers/2301.13826)

[Attend and Excite](../api/pipelines/stable_diffusion/attend_and_excite)를 사용하면 프롬프트의 피사체가 최종 이미지에 충실하게 표현되도록 할 수 있습니다.

이미지에 존재해야 하는 프롬프트의 피사체에 해당하는 일련의 토큰 인덱스가 입력으로 제공됩니다. 노이즈 제거 중에 각 토큰 인덱스는 이미지의 최소 한 패치 이상에 대해 최소 주의 임계값을 갖도록 보장됩니다. 모든 피사체 토큰에 대해 주의 임계값이 통과될 때까지 노이즈 제거 프로세스 중에 중간 잠복기가 반복적으로 최적화되어 가장 소홀히 취급되는 피사체 토큰의 주의력을 강화합니다.

Pix2Pix Zero와 마찬가지로 Attend and Excite 역시 파이프라인에 미니 최적화 루프(사전 학습된 가중치를 그대로 둔 채)가 포함되며, 일반적인 'StableDiffusionPipeline'보다 더 많은 메모리가 필요할 수 있습니다.

사용 방법에 대한 자세한 내용은 [여기](../api/pipelines/stable_diffusion/attend_and_excite)를 참조하세요.

## Semantic Guidance (SEGA)

[Paper](https://huggingface.co/papers/2301.12247)

의미유도(SEGA)를 사용하면 이미지에서 하나 이상의 컨셉을 적용하거나 제거할 수 있습니다. 컨셉의 강도도 조절할 수 있습니다. 즉, 스마일 컨셉을 사용하여 인물 사진의 스마일을 점진적으로 늘리거나 줄일 수 있습니다.

분류기 무료 안내(classifier free guidance)가 빈 프롬프트 입력을 통해 안내를 제공하는 방식과 유사하게, SEGA는 개념 프롬프트에 대한 안내를 제공합니다. 이러한 개념 프롬프트는 여러 개를 동시에 적용할 수 있습니다. 각 개념 프롬프트는 안내가 긍정적으로 적용되는지 또는 부정적으로 적용되는지에 따라 해당 개념을 추가하거나 제거할 수 있습니다.

Pix2Pix Zero 또는 Attend and Excite와 달리 SEGA는 명시적인 그라데이션 기반 최적화를 수행하는 대신 확산 프로세스와 직접 상호 작용합니다.

사용 방법에 대한 자세한 내용은 [여기](../api/pipelines/semantic_stable_diffusion)를 참조하세요.

## Self-attention Guidance (SAG)

[Paper](https://huggingface.co/papers/2210.00939)

[자기 주의 안내](../api/pipelines/stable_diffusion/self_attention_guidance)는 이미지의 전반적인 품질을 개선합니다.

SAG는 고빈도 세부 정보를 기반으로 하지 않은 예측에서 완전히 조건화된 이미지에 이르기까지 가이드를 제공합니다. 고빈도 디테일은 UNet 자기 주의 맵에서 추출됩니다.

사용 방법에 대한 자세한 내용은 [여기](../api/pipelines/stable_diffusion/self_attention_guidance)를 참조하세요.

## Depth2Image

[Project](https://huggingface.co/stabilityai/stable-diffusion-2-depth)

[Depth2Image](../pipelines/stable_diffusion_2#depthtoimage)는 텍스트 안내 이미지 변화에 대한 시맨틱을 더 잘 보존하도록 안정적 확산에서 미세 조정되었습니다.

원본 이미지의 단안(monocular) 깊이 추정치를 조건으로 합니다.

사용 방법에 대한 자세한 내용은 [여기](../api/pipelines/stable_diffusion_2#depthtoimage)를 참조하세요.

> [!TIP]
> InstructPix2Pix와 Pix2Pix Zero와 같은 방법의 중요한 차이점은 전자의 경우
> 는 사전 학습된 가중치를 미세 조정하는 반면, 후자는 그렇지 않다는 것입니다. 즉, 다음을 수행할 수 있습니다.
> 사용 가능한 모든 안정적 확산 모델에 Pix2Pix Zero를 적용할 수 있습니다.

## MultiDiffusion Panorama

[Paper](https://huggingface.co/papers/2302.08113)

MultiDiffusion은 사전 학습된 diffusion model을 통해 새로운 생성 프로세스를 정의합니다. 이 프로세스는 고품질의 다양한 이미지를 생성하는 데 쉽게 적용할 수 있는 여러 diffusion 생성 방법을 하나로 묶습니다. 결과는 원하는 종횡비(예: 파노라마) 및 타이트한 분할 마스크에서 바운딩 박스에 이르는 공간 안내 신호와 같은 사용자가 제공한 제어를 준수합니다.
[MultiDiffusion 파노라마](../api/pipelines/stable_diffusion/panorama)를 사용하면 임의의 종횡비(예: 파노라마)로 고품질 이미지를 생성할 수 있습니다.

파노라마 이미지를 생성하는 데 사용하는 방법에 대한 자세한 내용은 [여기](../api/pipelines/stable_diffusion/panorama)를 참조하세요.

## 나만의 모델 파인튜닝

사전 학습된 모델 외에도 Diffusers는 사용자가 제공한 데이터에 대해 모델을 파인튜닝할 수 있는 학습 스크립트가 있습니다.

## DreamBooth

[DreamBooth](../training/dreambooth)는 모델을 파인튜닝하여 새로운 주제에 대해 가르칩니다. 즉, 한 사람의 사진 몇 장을 사용하여 다양한 스타일로 그 사람의 이미지를 생성할 수 있습니다.

사용 방법에 대한 자세한 내용은 [여기](../training/dreambooth)를 참조하세요.

## Textual Inversion

[Textual Inversion](../training/text_inversion)은 모델을 파인튜닝하여 새로운 개념에 대해 학습시킵니다. 즉, 특정 스타일의 아트웍 사진 몇 장을 사용하여 해당 스타일의 이미지를 생성할 수 있습니다.

사용 방법에 대한 자세한 내용은 [여기](../training/text_inversion)를 참조하세요.

## ControlNet

[Paper](https://huggingface.co/papers/2302.05543)

[ControlNet](../api/pipelines/stable_diffusion/controlnet)은 추가 조건을 추가하는 보조 네트워크입니다.
가장자리 감지, 낙서, 깊이 맵, 의미적 세그먼트와 같은 다양한 조건에 대해 훈련된 8개의 표준 사전 훈련된 ControlNet이 있습니다,
깊이 맵, 시맨틱 세그먼테이션과 같은 다양한 조건으로 훈련된 8개의 표준 제어망이 있습니다.

사용 방법에 대한 자세한 내용은 [여기](../api/pipelines/stable_diffusion/controlnet)를 참조하세요.

## Prompt Weighting

프롬프트 가중치는 텍스트의 특정 부분에 더 많은 관심 가중치를 부여하는 간단한 기법입니다.
입력에 가중치를 부여하는 간단한 기법입니다.

자세한 설명과 예시는 [여기](../using-diffusers/weighted_prompts)를 참조하세요.

## Custom Diffusion

[Custom Diffusion](../training/custom_diffusion)은 사전 학습된 text-to-image 간 확산 모델의 교차 관심도 맵만 미세 조정합니다.
또한 textual inversion을 추가로 수행할 수 있습니다. 설계상 다중 개념 훈련을 지원합니다.
DreamBooth 및 Textual Inversion 마찬가지로, 사용자 지정 확산은 사전학습된 text-to-image diffusion 모델에 새로운 개념을 학습시켜 관심 있는 개념과 관련된 출력을 생성하는 데에도 사용됩니다.

자세한 설명은 [공식 문서](../training/custom_diffusion)를 참조하세요.

## Model Editing

[Paper](https://huggingface.co/papers/2303.08084)

[텍스트-이미지 모델 편집 파이프라인](../api/pipelines/model_editing)을 사용하면 사전학습된 text-to-image diffusion 모델이 입력 프롬프트에 있는 피사체에 대해 내릴 수 있는 잘못된 암시적 가정을 완화하는 데 도움이 됩니다.
예를 들어, 안정적 확산에 "A pack of roses"에 대한 이미지를 생성하라는 메시지를 표시하면 생성된 이미지의 장미는 빨간색일 가능성이 높습니다. 이 파이프라인은 이러한 가정을 변경하는 데 도움이 됩니다.

자세한 설명은 [공식 문서](../api/pipelines/model_editing)를 참조하세요.

## DiffEdit

[Paper](https://huggingface.co/papers/2210.11427)

[DiffEdit](../api/pipelines/diffedit)를 사용하면 원본 입력 이미지를 최대한 보존하면서 입력 프롬프트와 함께 입력 이미지의 의미론적 편집이 가능합니다.


자세한 설명은 [공식 문서](../api/pipelines/diffedit)를 참조하세요.

## T2I-Adapter

[Paper](https://huggingface.co/papers/2302.08453)

[T2I-어댑터](../api/pipelines/stable_diffusion/adapter)는 추가적인 조건을 추가하는 auxiliary 네트워크입니다.
가장자리 감지, 스케치, depth maps, semantic segmentations와 같은 다양한 조건에 대해 훈련된 8개의 표준 사전훈련된 adapter가 있습니다,

[공식 문서](api/pipelines/stable_diffusion/adapter)에서 사용 방법에 대한 정보를 참조하세요.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/controlling_generation.md" />

### 파이프라인, 모델 및 스케줄러 이해하기
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/write_own_pipeline.md

# 파이프라인, 모델 및 스케줄러 이해하기


🧨 Diffusers는 사용자 친화적이며 유연한 도구 상자로, 사용사례에 맞게 diffusion 시스템을 구축 할 수 있도록 설계되었습니다. 이 도구 상자의 핵심은 모델과 스케줄러입니다. `DiffusionPipeline`은 편의를 위해 이러한 구성 요소를 번들로 제공하지만, 파이프라인을 분리하고 모델과 스케줄러를 개별적으로 사용해 새로운 diffusion 시스템을 만들 수도 있습니다.

이 튜토리얼에서는 기본 파이프라인부터 시작해 Stable Diffusion 파이프라인까지 진행하며 모델과 스케줄러를 사용해 추론을 위한 diffusion 시스템을 조립하는 방법을 배웁니다.

## 기본 파이프라인 해체하기

파이프라인은 추론을 위해 모델을 실행하는 빠르고 쉬운 방법으로, 이미지를 생성하는 데 코드가 4줄 이상 필요하지 않습니다:

```py
>>> from diffusers import DDPMPipeline

>>> ddpm = DDPMPipeline.from_pretrained("google/ddpm-cat-256").to("cuda")
>>> image = ddpm(num_inference_steps=25).images[0]
>>> image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/ddpm-cat.png" alt="Image of cat created from DDPMPipeline"/>
</div>

정말 쉽습니다. 그런데 파이프라인은 어떻게 이렇게 할 수 있었을까요? 파이프라인을 세분화하여 내부에서 어떤 일이 일어나고 있는지 살펴보겠습니다.

위 예시에서 파이프라인에는 `UNet2DModel` 모델과 `DDPMScheduler`가 포함되어 있습니다. 파이프라인은 원하는 출력 크기의 랜덤 노이즈를 받아 모델을 여러번 통과시켜 이미지의 노이즈를 제거합니다. 각 timestep에서 모델은 *noise residual*을 예측하고 스케줄러는 이를 사용하여 노이즈가 적은 이미지를 예측합니다. 파이프라인은 지정된 추론 스텝수에 도달할 때까지 이 과정을 반복합니다.

모델과 스케줄러를 별도로 사용하여 파이프라인을 다시 생성하기 위해 자체적인 노이즈 제거 프로세스를 작성해 보겠습니다.

1. 모델과 스케줄러를 불러옵니다:

    ```py
    >>> from diffusers import DDPMScheduler, UNet2DModel

    >>> scheduler = DDPMScheduler.from_pretrained("google/ddpm-cat-256")
    >>> model = UNet2DModel.from_pretrained("google/ddpm-cat-256").to("cuda")
    ```

2. 노이즈 제거 프로세스를 실행할 timestep 수를 설정합니다:

    ```py
    >>> scheduler.set_timesteps(50)
    ```

3. 스케줄러의 timestep을 설정하면 균등한 간격의 구성 요소를 가진 텐서가 생성됩니다.(이 예시에서는 50개) 각 요소는 모델이 이미지의 노이즈를 제거하는 시간 간격에 해당합니다. 나중에 노이즈 제거 루프를 만들 때 이 텐서를 반복하여 이미지의 노이즈를 제거합니다:

    ```py
    >>> scheduler.timesteps
    tensor([980, 960, 940, 920, 900, 880, 860, 840, 820, 800, 780, 760, 740, 720,
        700, 680, 660, 640, 620, 600, 580, 560, 540, 520, 500, 480, 460, 440,
        420, 400, 380, 360, 340, 320, 300, 280, 260, 240, 220, 200, 180, 160,
        140, 120, 100,  80,  60,  40,  20,   0])
    ```

4. 원하는 출력과 같은 모양을 가진 랜덤 노이즈를 생성합니다:

    ```py
    >>> import torch

    >>> sample_size = model.config.sample_size
    >>> noise = torch.randn((1, 3, sample_size, sample_size), device="cuda")
    ```

5. 이제 timestep을 반복하는 루프를 작성합니다. 각 timestep에서 모델은 `UNet2DModel.forward()`를 통해 noisy residual을 반환합니다. 스케줄러의 `step()` 메서드는 noisy residual, timestep, 그리고 입력을 받아 이전 timestep에서 이미지를 예측합니다. 이 출력은 노이즈 제거 루프의 모델에 대한 다음 입력이 되며, `timesteps` 배열의 끝에 도달할 때까지 반복됩니다.

    ```py
    >>> input = noise

    >>> for t in scheduler.timesteps:
    ...     with torch.no_grad():
    ...         noisy_residual = model(input, t).sample
    ...     previous_noisy_sample = scheduler.step(noisy_residual, t, input).prev_sample
    ...     input = previous_noisy_sample
    ```

    이것이 전체 노이즈 제거 프로세스이며, 동일한 패턴을 사용해 모든 diffusion 시스템을 작성할 수 있습니다.

6. 마지막 단계는 노이즈가 제거된 출력을 이미지로 변환하는 것입니다:

    ```py
    >>> from PIL import Image
    >>> import numpy as np

    >>> image = (input / 2 + 0.5).clamp(0, 1)
    >>> image = image.cpu().permute(0, 2, 3, 1).numpy()[0]
    >>> image = Image.fromarray((image * 255).round().astype("uint8"))
    >>> image
    ```

다음 섹션에서는 여러분의 기술을 시험해보고 좀 더 복잡한 Stable Diffusion 파이프라인을 분석해 보겠습니다. 방법은 거의 동일합니다. 필요한 구성요소들을 초기화하고 timestep수를 설정하여 `timestep` 배열을 생성합니다. 노이즈 제거 루프에서 `timestep` 배열이 사용되며, 이 배열의 각 요소에 대해 모델은 노이즈가 적은 이미지를 예측합니다. 노이즈 제거 루프는 `timestep`을 반복하고 각 timestep에서 noise residual을 출력하고 스케줄러는 이를 사용하여 이전 timestep에서 노이즈가 덜한 이미지를 예측합니다. 이 프로세스는 `timestep` 배열의 끝에 도달할 때까지 반복됩니다.

한번 사용해 봅시다!

## Stable Diffusion 파이프라인 해체하기

Stable Diffusion 은 text-to-image *latent diffusion* 모델입니다. latent diffusion 모델이라고 불리는 이유는 실제 픽셀 공간 대신 이미지의 저차원의 표현으로 작업하기 때문이고, 메모리 효율이 더 높습니다. 인코더는 이미지를 더 작은 표현으로 압축하고, 디코더는 압축된 표현을 다시 이미지로 변환합니다. text-to-image 모델의 경우 텍스트 임베딩을 생성하기 위해 tokenizer와 인코더가 필요합니다. 이전 예제에서 이미 UNet 모델과 스케줄러가 필요하다는 것은 알고 계셨을 것입니다.

보시다시피, 이것은 UNet 모델만 포함된 DDPM 파이프라인보다 더 복잡합니다. Stable Diffusion 모델에는 세 개의 개별 사전학습된 모델이 있습니다.

> [!TIP]
> 💡 VAE, UNet 및 텍스트 인코더 모델의 작동방식에 대한 자세한 내용은 [How does Stable Diffusion work?](https://huggingface.co/blog/stable_diffusion#how-does-stable-diffusion-work) 블로그를 참조하세요.

이제 Stable Diffusion 파이프라인에 필요한 구성요소들이 무엇인지 알았으니, `from_pretrained()` 메서드를 사용해 모든 구성요소를 불러옵니다. 사전학습된 체크포인트 [`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)에서 찾을 수 있으며, 각 구성요소들은 별도의 하위 폴더에 저장되어 있습니다:

```py
>>> from PIL import Image
>>> import torch
>>> from transformers import CLIPTextModel, CLIPTokenizer
>>> from diffusers import AutoencoderKL, UNet2DConditionModel, PNDMScheduler

>>> vae = AutoencoderKL.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="vae")
>>> tokenizer = CLIPTokenizer.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="tokenizer")
>>> text_encoder = CLIPTextModel.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="text_encoder")
>>> unet = UNet2DConditionModel.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="unet")
```

기본 `PNDMScheduler` 대신, `UniPCMultistepScheduler`로 교체하여 다른 스케줄러를 얼마나 쉽게 연결할 수 있는지 확인합니다:

```py
>>> from diffusers import UniPCMultistepScheduler

>>> scheduler = UniPCMultistepScheduler.from_pretrained("CompVis/stable-diffusion-v1-4", subfolder="scheduler")
```

추론 속도를 높이려면 스케줄러와 달리 학습 가능한 가중치가 있으므로 모델을 GPU로 옮기세요:

```py
>>> torch_device = "cuda"
>>> vae.to(torch_device)
>>> text_encoder.to(torch_device)
>>> unet.to(torch_device)
```

### 텍스트 임베딩 생성하기

다음 단계는 임베딩을 생성하기 위해 텍스트를 토큰화하는 것입니다. 이 텍스트는 UNet 모델에서 condition으로 사용되고 입력 프롬프트와 유사한 방향으로 diffusion 프로세스를 조정하는 데 사용됩니다.

> [!TIP]
> 💡 `guidance_scale` 매개변수는 이미지를 생성할 때 프롬프트에 얼마나 많은 가중치를 부여할지 결정합니다.

다른 프롬프트를 생성하고 싶다면 원하는 프롬프트를 자유롭게 선택하세요!

```py
>>> prompt = ["a photograph of an astronaut riding a horse"]
>>> height = 512  # Stable Diffusion의 기본 높이
>>> width = 512  # Stable Diffusion의 기본 너비
>>> num_inference_steps = 25  # 노이즈 제거 스텝 수
>>> guidance_scale = 7.5  # classifier-free guidance를 위한 scale
>>> generator = torch.manual_seed(0)  # 초기 잠재 노이즈를 생성하는 seed generator
>>> batch_size = len(prompt)
```

텍스트를 토큰화하고 프롬프트에서 임베딩을 생성합니다:

```py
>>> text_input = tokenizer(
...     prompt, padding="max_length", max_length=tokenizer.model_max_length, truncation=True, return_tensors="pt"
... )

>>> with torch.no_grad():
...     text_embeddings = text_encoder(text_input.input_ids.to(torch_device))[0]
```

또한 패딩 토큰의 임베딩인 *unconditional 텍스트 임베딩*을 생성해야 합니다. 이 임베딩은 조건부 `text_embeddings`과 동일한 shape(`batch_size` 그리고 `seq_length`)을 가져야 합니다:

```py
>>> max_length = text_input.input_ids.shape[-1]
>>> uncond_input = tokenizer([""] * batch_size, padding="max_length", max_length=max_length, return_tensors="pt")
>>> uncond_embeddings = text_encoder(uncond_input.input_ids.to(torch_device))[0]
```

두번의 forward pass를 피하기 위해 conditional 임베딩과 unconditional 임베딩을 배치(batch)로 연결하겠습니다:

```py
>>> text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
```

### 랜덤 노이즈 생성

그다음 diffusion 프로세스의 시작점으로 초기 랜덤 노이즈를 생성합니다. 이것이 이미지의 잠재적 표현이며 점차적으로 노이즈가 제거됩니다. 이 시점에서 `latent` 이미지는 최종 이미지 크기보다 작지만 나중에 모델이 이를 512x512 이미지 크기로 변환하므로 괜찮습니다.

> [!TIP]
> 💡 `vae` 모델에는 3개의 다운 샘플링 레이어가 있기 때문에 높이와 너비가 8로 나뉩니다. 다음을 실행하여 확인할 수 있습니다:
>
> ```py
> 2 ** (len(vae.config.block_out_channels) - 1) == 8
> ```

```py
>>> latents = torch.randn(
...     (batch_size, unet.config.in_channels, height // 8, width // 8),
...     generator=generator,
...     device=torch_device,
... )
```

### 이미지 노이즈 제거

먼저 `UniPCMultistepScheduler`와 같은 향상된 스케줄러에 필요한 노이즈 스케일 값인 초기 노이즈 분포 *sigma* 로 입력을 스케일링 하는 것부터 시작합니다:

```py
>>> latents = latents * scheduler.init_noise_sigma
```

마지막 단계는 `latent`의 순수한 노이즈를 점진적으로 프롬프트에 설명된 이미지로 변환하는 노이즈 제거 루프를 생성하는 것입니다. 노이즈 제거 루프는 세 가지 작업을 수행해야 한다는 점을 기억하세요:

1. 노이즈 제거 중에 사용할 스케줄러의 timesteps를 설정합니다.
2. timestep을 따라 반복합니다.
3. 각 timestep에서 UNet 모델을 호출하여 noise residual을 예측하고 스케줄러에 전달하여 이전 노이즈 샘플을 계산합니다.

```py
>>> from tqdm.auto import tqdm

>>> scheduler.set_timesteps(num_inference_steps)

>>> for t in tqdm(scheduler.timesteps):
...     # classifier-free guidance를 수행하는 경우 두번의 forward pass를 수행하지 않도록 latent를 확장.
...     latent_model_input = torch.cat([latents] * 2)

...     latent_model_input = scheduler.scale_model_input(latent_model_input, timestep=t)

...     # noise residual 예측
...     with torch.no_grad():
...         noise_pred = unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample

...     # guidance 수행
...     noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
...     noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

...     # 이전 노이즈 샘플을 계산 x_t -> x_t-1
...     latents = scheduler.step(noise_pred, t, latents).prev_sample
```

### 이미지 디코딩

마지막 단계는 `vae`를 이용하여 잠재 표현을 이미지로 디코딩하고 `sample`과 함께 디코딩된 출력을 얻는 것입니다:

```py
# latent를 스케일링하고 vae로 이미지 디코딩
latents = 1 / 0.18215 * latents
with torch.no_grad():
    image = vae.decode(latents).sample
```

마지막으로 이미지를 `PIL.Image`로 변환하면 생성된 이미지를 확인할 수 있습니다!

```py
>>> image = (image / 2 + 0.5).clamp(0, 1)
>>> image = image.detach().cpu().permute(0, 2, 3, 1).numpy()
>>> images = (image * 255).round().astype("uint8")
>>> pil_images = [Image.fromarray(image) for image in images]
>>> pil_images[0]
```

<div class="flex justify-center">
    <img src="https://huggingface.co/blog/assets/98_stable_diffusion/stable_diffusion_k_lms.png"/>
</div>

## 다음 단계

기본 파이프라인부터 복잡한 파이프라인까지, 자신만의 diffusion 시스템을 작성하는 데 필요한 것은 노이즈 제거 루프뿐이라는 것을 알 수 있었습니다. 이 루프는 스케줄러의 timesteps를 설정하고, 이를 반복하며, UNet 모델을 호출하여 noise residual을 예측하고 스케줄러에 전달하여 이전 노이즈 샘플을 계산하는 과정을 번갈아 가며 수행해야 합니다.

이것이 바로 🧨 Diffusers가 설계된 목적입니다: 모델과 스케줄러를 사용해 자신만의 diffusion 시스템을 직관적이고 쉽게 작성할 수 있도록 하기 위해서입니다.

다음 단계를 자유롭게 진행하세요:

* 🧨 Diffusers에 [파이프라인 구축 및 기여](using-diffusers/#contribute_pipeline)하는 방법을 알아보세요. 여러분이 어떤 아이디어를 내놓을지 기대됩니다!
* 라이브러리에서 [기본 파이프라인](./api/pipelines/overview)을 살펴보고, 모델과 스케줄러를 별도로 사용하여 파이프라인을 처음부터 해체하고 빌드할 수 있는지 확인해 보세요.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/write_own_pipeline.md" />

### JAX / Flax에서의 🧨 Stable Diffusion!
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/stable_diffusion_jax_how_to.md

# JAX / Flax에서의 🧨 Stable Diffusion!


🤗 Hugging Face [Diffusers] (https://github.com/huggingface/diffusers) 는 버전 0.5.1부터 Flax를 지원합니다! 이를 통해 Colab, Kaggle, Google Cloud Platform에서 사용할 수 있는 것처럼 Google TPU에서 초고속 추론이 가능합니다.

이 노트북은 JAX / Flax를 사용해 추론을 실행하는 방법을 보여줍니다. Stable Diffusion의 작동 방식에 대한 자세한 내용을 원하거나 GPU에서 실행하려면 이 [노트북] ](https://huggingface.co/docs/diffusers/stable_diffusion)을 참조하세요.

먼저, TPU 백엔드를 사용하고 있는지 확인합니다. Colab에서 이 노트북을 실행하는 경우, 메뉴에서 런타임을 선택한 다음 "런타임 유형 변경" 옵션을 선택한 다음 하드웨어 가속기 설정에서 TPU를 선택합니다.

JAX는 TPU 전용은 아니지만 각 TPU 서버에는 8개의 TPU 가속기가 병렬로 작동하기 때문에 해당 하드웨어에서 더 빛을 발한다는 점은 알아두세요.


## Setup

먼저 diffusers가 설치되어 있는지 확인합니다.

```bash
!pip install jax==0.3.25 jaxlib==0.3.25 flax transformers ftfy
!pip install diffusers
```

```python
import jax.tools.colab_tpu

jax.tools.colab_tpu.setup_tpu()
import jax
```

```python
num_devices = jax.device_count()
device_type = jax.devices()[0].device_kind

print(f"Found {num_devices} JAX devices of type {device_type}.")
assert (
    "TPU" in device_type
), "Available device is not a TPU, please select TPU from Edit > Notebook settings > Hardware accelerator"
```

```python out
Found 8 JAX devices of type Cloud TPU.
```

그런 다음 모든 dependencies를 가져옵니다.

```python
import numpy as np
import jax
import jax.numpy as jnp

from pathlib import Path
from jax import pmap
from flax.jax_utils import replicate
from flax.training.common_utils import shard
from PIL import Image

from huggingface_hub import notebook_login
from diffusers import FlaxStableDiffusionPipeline
```

## 모델 불러오기

TPU 장치는 효율적인 half-float 유형인 bfloat16을 지원합니다. 테스트에는 이 유형을 사용하지만 대신 float32를 사용하여 전체 정밀도(full precision)를 사용할 수도 있습니다.

```python
dtype = jnp.bfloat16
```

Flax는 함수형 프레임워크이므로 모델은 무상태(stateless)형이며 매개변수는 모델 외부에 저장됩니다. 사전학습된 Flax 파이프라인을 불러오면 파이프라인 자체와 모델 가중치(또는 매개변수)가 모두 반환됩니다. 저희는 bf16 버전의 가중치를 사용하고 있으므로 유형 경고가 표시되지만 무시해도 됩니다.

```python
pipeline, params = FlaxStableDiffusionPipeline.from_pretrained(
    "CompVis/stable-diffusion-v1-4",
    variant="bf16",
    dtype=dtype,
)
```

## 추론

TPU에는 일반적으로 8개의 디바이스가 병렬로 작동하므로 보유한 디바이스 수만큼 프롬프트를 복제합니다. 그런 다음 각각 하나의 이미지 생성을 담당하는 8개의 디바이스에서 한 번에 추론을 수행합니다. 따라서 하나의 칩이 하나의 이미지를 생성하는 데 걸리는 시간과 동일한 시간에 8개의 이미지를 얻을 수 있습니다.

프롬프트를 복제하고 나면 파이프라인의 `prepare_inputs` 함수를 호출하여 토큰화된 텍스트 ID를 얻습니다. 토큰화된 텍스트의 길이는 기본 CLIP 텍스트 모델의 구성에 따라 77토큰으로 설정됩니다.

```python
prompt = "A cinematic film still of Morgan Freeman starring as Jimi Hendrix, portrait, 40mm lens, shallow depth of field, close up, split lighting, cinematic"
prompt = [prompt] * jax.device_count()
prompt_ids = pipeline.prepare_inputs(prompt)
prompt_ids.shape
```

```python out
(8, 77)
```

### 복사(Replication) 및 정렬화

모델 매개변수와 입력값은 우리가 보유한 8개의 병렬 장치에 복사(Replication)되어야 합니다. 매개변수 딕셔너리는 `flax.jax_utils.replicate`(딕셔너리를 순회하며 가중치의 모양을 변경하여 8번 반복하는 함수)를 사용하여 복사됩니다. 배열은 `shard`를 사용하여 복제됩니다.

```python
p_params = replicate(params)
```

```python
prompt_ids = shard(prompt_ids)
prompt_ids.shape
```

```python out
(8, 1, 77)
```

이 shape은 8개의 디바이스 각각이 shape `(1, 77)`의 jnp 배열을 입력값으로 받는다는 의미입니다. 즉 1은 디바이스당 batch(배치) 크기입니다. 메모리가 충분한 TPU에서는 한 번에 여러 이미지(칩당)를 생성하려는 경우 1보다 클 수 있습니다.

이미지를 생성할 준비가 거의 완료되었습니다! 이제 생성 함수에 전달할 난수 생성기만 만들면 됩니다. 이것은 난수를 다루는 모든 함수에 난수 생성기가 있어야 한다는, 난수에 대해 매우 진지하고 독단적인 Flax의 표준 절차입니다. 이렇게 하면 여러 분산된 기기에서 훈련할 때에도 재현성이 보장됩니다.

아래 헬퍼 함수는 시드를 사용하여 난수 생성기를 초기화합니다. 동일한 시드를 사용하는 한 정확히 동일한 결과를 얻을 수 있습니다. 나중에 노트북에서 결과를 탐색할 때엔 다른 시드를 자유롭게 사용하세요.

```python
def create_key(seed=0):
    return jax.random.PRNGKey(seed)
```

rng를 얻은 다음 8번 '분할'하여 각 디바이스가 다른 제너레이터를 수신하도록 합니다. 따라서 각 디바이스마다 다른 이미지가 생성되며 전체 프로세스를 재현할 수 있습니다.

```python
rng = create_key(0)
rng = jax.random.split(rng, jax.device_count())
```

JAX 코드는 매우 빠르게 실행되는 효율적인 표현으로 컴파일할 수 있습니다. 하지만 후속 호출에서 모든 입력이 동일한 모양을 갖도록 해야 하며, 그렇지 않으면 JAX가 코드를 다시 컴파일해야 하므로 최적화된 속도를 활용할 수 없습니다.

`jit = True`를 인수로 전달하면 Flax 파이프라인이 코드를 컴파일할 수 있습니다. 또한 모델이 사용 가능한 8개의 디바이스에서 병렬로 실행되도록 보장합니다.

다음 셀을 처음 실행하면 컴파일하는 데 시간이 오래 걸리지만 이후 호출(입력이 다른 경우에도)은 훨씬 빨라집니다. 예를 들어, 테스트했을 때 TPU v2-8에서 컴파일하는 데 1분 이상 걸리지만 이후 추론 실행에는 약 7초가 걸립니다.

```
%%time
images = pipeline(prompt_ids, p_params, rng, jit=True)[0]
```

```python out
CPU times: user 56.2 s, sys: 42.5 s, total: 1min 38s
Wall time: 1min 29s
```

반환된 배열의 shape은 `(8, 1, 512, 512, 3)`입니다. 이를 재구성하여 두 번째 차원을 제거하고 512 × 512 × 3의 이미지 8개를 얻은 다음 PIL로 변환합니다.

```python
images = images.reshape((images.shape[0] * images.shape[1],) + images.shape[-3:])
images = pipeline.numpy_to_pil(images)
```

### 시각화

이미지를 그리드에 표시하는 도우미 함수를 만들어 보겠습니다.

```python
def image_grid(imgs, rows, cols):
    w, h = imgs[0].size
    grid = Image.new("RGB", size=(cols * w, rows * h))
    for i, img in enumerate(imgs):
        grid.paste(img, box=(i % cols * w, i // cols * h))
    return grid
```

```python
image_grid(images, 2, 4)
```

![img](https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/stable_diffusion_jax_how_to_cell_38_output_0.jpeg)


## 다른 프롬프트 사용

모든 디바이스에서 동일한 프롬프트를 복제할 필요는 없습니다. 프롬프트 2개를 각각 4번씩 생성하거나 한 번에 8개의 서로 다른 프롬프트를 생성하는 등 원하는 것은 무엇이든 할 수 있습니다. 한번 해보세요!

먼저 입력 준비 코드를 편리한 함수로 리팩터링하겠습니다:

```python
prompts = [
    "Labrador in the style of Hokusai",
    "Painting of a squirrel skating in New York",
    "HAL-9000 in the style of Van Gogh",
    "Times Square under water, with fish and a dolphin swimming around",
    "Ancient Roman fresco showing a man working on his laptop",
    "Close-up photograph of young black woman against urban background, high quality, bokeh",
    "Armchair in the shape of an avocado",
    "Clown astronaut in space, with Earth in the background",
]
```

```python
prompt_ids = pipeline.prepare_inputs(prompts)
prompt_ids = shard(prompt_ids)

images = pipeline(prompt_ids, p_params, rng, jit=True).images
images = images.reshape((images.shape[0] * images.shape[1],) + images.shape[-3:])
images = pipeline.numpy_to_pil(images)

image_grid(images, 2, 4)
```

![img](https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/stable_diffusion_jax_how_to_cell_43_output_0.jpeg)


## 병렬화(parallelization)는 어떻게 작동하는가?

앞서 `diffusers` Flax 파이프라인이 모델을 자동으로 컴파일하고 사용 가능한 모든 기기에서 병렬로 실행한다고 말씀드렸습니다. 이제 그 프로세스를 간략하게 살펴보고 작동 방식을 보여드리겠습니다.

JAX 병렬화는 여러 가지 방법으로 수행할 수 있습니다. 가장 쉬운 방법은 jax.pmap 함수를 사용하여 단일 프로그램, 다중 데이터(SPMD) 병렬화를 달성하는 것입니다. 즉, 동일한 코드의 복사본을 각각 다른 데이터 입력에 대해 여러 개 실행하는 것입니다. 더 정교한 접근 방식도 가능하므로 관심이 있으시다면 [JAX 문서](https://jax.readthedocs.io/en/latest/index.html)와 [`pjit` 페이지](https://jax.readthedocs.io/en/latest/jax-101/08-pjit.html?highlight=pjit)에서 이 주제를 살펴보시기 바랍니다!

`jax.pmap`은 두 가지 기능을 수행합니다:

- `jax.jit()`를 호출한 것처럼 코드를 컴파일(또는 `jit`)합니다. 이 작업은 `pmap`을 호출할 때가 아니라 pmapped 함수가 처음 호출될 때 수행됩니다.
- 컴파일된 코드가 사용 가능한 모든 기기에서 병렬로 실행되도록 합니다.

작동 방식을 보여드리기 위해 이미지 생성을 실행하는 비공개 메서드인 파이프라인의 `_generate` 메서드를 `pmap`합니다. 이 메서드는 향후 `Diffusers` 릴리스에서 이름이 변경되거나 제거될 수 있다는 점에 유의하세요.

```python
p_generate = pmap(pipeline._generate)
```

`pmap`을 사용한 후 준비된 함수 `p_generate`는 개념적으로 다음을 수행합니다:
* 각 장치에서 기본 함수 `pipeline._generate`의 복사본을 호출합니다.
* 각 장치에 입력 인수의 다른 부분을 보냅니다. 이것이 바로 샤딩이 사용되는 이유입니다. 이 경우 `prompt_ids`의 shape은 `(8, 1, 77, 768)`입니다. 이 배열은 8개로 분할되고 `_generate`의 각 복사본은 `(1, 77, 768)`의 shape을 가진 입력을 받게 됩니다.

병렬로 호출된다는 사실을 완전히 무시하고 `_generate`를 코딩할 수 있습니다. batch(배치) 크기(이 예제에서는 `1`)와 코드에 적합한 차원만 신경 쓰면 되며, 병렬로 작동하기 위해 아무것도 변경할 필요가 없습니다.

파이프라인 호출을 사용할 때와 마찬가지로, 다음 셀을 처음 실행할 때는 시간이 걸리지만 그 이후에는 훨씬 빨라집니다.

```
%%time
images = p_generate(prompt_ids, p_params, rng)
images = images.block_until_ready()
images.shape
```

```python out
CPU times: user 1min 15s, sys: 18.2 s, total: 1min 34s
Wall time: 1min 15s
```

```python
images.shape
```

```python out
(8, 1, 512, 512, 3)
```

JAX는 비동기 디스패치를 사용하고 가능한 한 빨리 제어권을 Python 루프에 반환하기 때문에 추론 시간을 정확하게 측정하기 위해 `block_until_ready()`를 사용합니다. 아직 구체화되지 않은 계산 결과를 사용하려는 경우 자동으로 차단이 수행되므로 코드에서 이 함수를 사용할 필요가 없습니다.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/stable_diffusion_jax_how_to.md" />

### DiffEdit
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/diffedit.md

# DiffEdit


이미지 편집을 하려면 일반적으로 편집할 영역의 마스크를 제공해야 합니다. DiffEdit는 텍스트 쿼리를 기반으로 마스크를 자동으로 생성하므로 이미지 편집 소프트웨어 없이도 마스크를 만들기가 전반적으로 더 쉬워집니다. DiffEdit 알고리즘은 세 단계로 작동합니다:

1. Diffusion 모델이 일부 쿼리 텍스트와 참조 텍스트를 조건부로 이미지의 노이즈를 제거하여 이미지의 여러 영역에 대해 서로 다른 노이즈 추정치를 생성하고, 그 차이를 사용하여 쿼리 텍스트와 일치하도록 이미지의 어느 영역을 변경해야 하는지 식별하기 위한 마스크를 추론합니다.
2. 입력 이미지가 DDIM을 사용하여 잠재 공간으로 인코딩됩니다.
3. 마스크 외부의 픽셀이 입력 이미지와 동일하게 유지되도록 마스크를 가이드로 사용하여 텍스트 쿼리에 조건이 지정된 diffusion 모델로 latents를 디코딩합니다.

이 가이드에서는 마스크를 수동으로 만들지 않고 DiffEdit를 사용하여 이미지를 편집하는 방법을 설명합니다.

시작하기 전에 다음 라이브러리가 설치되어 있는지 확인하세요:

```py
# Colab에서 필요한 라이브러리를 설치하기 위해 주석을 제외하세요
#!pip install -q diffusers transformers accelerate
```

`StableDiffusionDiffEditPipeline`에는 이미지 마스크와 부분적으로 반전된 latents 집합이 필요합니다. 이미지 마스크는 `generate_mask()` 함수에서 생성되며, 두 개의 파라미터인 `source_prompt`와 `target_prompt`가 포함됩니다. 이 매개변수는 이미지에서 무엇을 편집할지 결정합니다. 예를 들어, *과일* 한 그릇을 *배* 한 그릇으로 변경하려면 다음과 같이 하세요:

```py
source_prompt = "a bowl of fruits"
target_prompt = "a bowl of pears"
```

부분적으로 반전된 latents는 `invert()` 함수에서 생성되며, 일반적으로 이미지를 설명하는 `prompt` 또는 *캡션*을 포함하는 것이 inverse latent sampling 프로세스를 가이드하는 데 도움이 됩니다. 캡션은 종종 `source_prompt`가 될 수 있지만, 다른 텍스트 설명으로 자유롭게 실험해 보세요!

파이프라인, 스케줄러, 역 스케줄러를 불러오고 메모리 사용량을 줄이기 위해 몇 가지 최적화를 활성화해 보겠습니다:

```py
import torch
from diffusers import DDIMScheduler, DDIMInverseScheduler, StableDiffusionDiffEditPipeline

pipeline = StableDiffusionDiffEditPipeline.from_pretrained(
    "stabilityai/stable-diffusion-2-1",
    torch_dtype=torch.float16,
    safety_checker=None,
    use_safetensors=True,
)
pipeline.scheduler = DDIMScheduler.from_config(pipeline.scheduler.config)
pipeline.inverse_scheduler = DDIMInverseScheduler.from_config(pipeline.scheduler.config)
pipeline.enable_model_cpu_offload()
pipeline.enable_vae_slicing()
```

수정하기 위한 이미지를 불러옵니다:

```py
from diffusers.utils import load_image, make_image_grid

img_url = "https://github.com/Xiang-cd/DiffEdit-stable-diffusion/raw/main/assets/origin.png"
raw_image = load_image(img_url).resize((768, 768))
raw_image
```

이미지 마스크를 생성하기 위해 `generate_mask()` 함수를 사용합니다. 이미지에서 편집할 내용을 지정하기 위해 `source_prompt`와 `target_prompt`를 전달해야 합니다:

```py
from PIL import Image

source_prompt = "a bowl of fruits"
target_prompt = "a basket of pears"
mask_image = pipeline.generate_mask(
    image=raw_image,
    source_prompt=source_prompt,
    target_prompt=target_prompt,
)
Image.fromarray((mask_image.squeeze()*255).astype("uint8"), "L").resize((768, 768))
```

다음으로, 반전된 latents를 생성하고 이미지를 묘사하는 캡션에 전달합니다:

```py
inv_latents = pipeline.invert(prompt=source_prompt, image=raw_image).latents
```

마지막으로, 이미지 마스크와 반전된 latents를 파이프라인에 전달합니다. `target_prompt`는 이제 `prompt`가 되며, `source_prompt`는 `negative_prompt`로 사용됩니다.

```py
output_image = pipeline(
    prompt=target_prompt,
    mask_image=mask_image,
    image_latents=inv_latents,
    negative_prompt=source_prompt,
).images[0]
mask_image = Image.fromarray((mask_image.squeeze()*255).astype("uint8"), "L").resize((768, 768))
make_image_grid([raw_image, mask_image, output_image], rows=1, cols=3)
```

<div class="flex gap-4">
  <div>
    <img class="rounded-xl" src="https://github.com/Xiang-cd/DiffEdit-stable-diffusion/raw/main/assets/origin.png"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">original image</figcaption>
  </div>
  <div>
    <img class="rounded-xl" src="https://github.com/Xiang-cd/DiffEdit-stable-diffusion/blob/main/assets/target.png?raw=true"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">edited image</figcaption>
  </div>
</div>

## Source와 target 임베딩 생성하기

Source와 target 임베딩은 수동으로 생성하는 대신 [Flan-T5](https://huggingface.co/docs/transformers/model_doc/flan-t5) 모델을 사용하여 자동으로 생성할 수 있습니다.

Flan-T5 모델과 토크나이저를 🤗 Transformers 라이브러리에서 불러옵니다:

```py
import torch
from transformers import AutoTokenizer, T5ForConditionalGeneration

tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-large")
model = T5ForConditionalGeneration.from_pretrained("google/flan-t5-large", device_map="auto", torch_dtype=torch.float16)
```

모델에 프롬프트할 source와 target 프롬프트를 생성하기 위해 초기 텍스트들을 제공합니다.

```py
source_concept = "bowl"
target_concept = "basket"

source_text = f"Provide a caption for images containing a {source_concept}. "
"The captions should be in English and should be no longer than 150 characters."

target_text = f"Provide a caption for images containing a {target_concept}. "
"The captions should be in English and should be no longer than 150 characters."
```

다음으로, 프롬프트들을 생성하기 위해 유틸리티 함수를 생성합니다.

```py
@torch.no_grad()
def generate_prompts(input_prompt):
    input_ids = tokenizer(input_prompt, return_tensors="pt").input_ids.to("cuda")

    outputs = model.generate(
        input_ids, temperature=0.8, num_return_sequences=16, do_sample=True, max_new_tokens=128, top_k=10
    )
    return tokenizer.batch_decode(outputs, skip_special_tokens=True)

source_prompts = generate_prompts(source_text)
target_prompts = generate_prompts(target_text)
print(source_prompts)
print(target_prompts)
```

> [!TIP]
> 다양한 품질의 텍스트를 생성하는 전략에 대해 자세히 알아보려면 [생성 전략](https://huggingface.co/docs/transformers/main/en/generation_strategies) 가이드를 참조하세요.

텍스트 인코딩을 위해 `StableDiffusionDiffEditPipeline`에서 사용하는 텍스트 인코더 모델을 불러옵니다. 텍스트 인코더를 사용하여 텍스트 임베딩을 계산합니다:

```py
import torch
from diffusers import StableDiffusionDiffEditPipeline

pipeline = StableDiffusionDiffEditPipeline.from_pretrained(
    "stabilityai/stable-diffusion-2-1", torch_dtype=torch.float16, use_safetensors=True
)
pipeline.enable_model_cpu_offload()
pipeline.enable_vae_slicing()

@torch.no_grad()
def embed_prompts(sentences, tokenizer, text_encoder, device="cuda"):
    embeddings = []
    for sent in sentences:
        text_inputs = tokenizer(
            sent,
            padding="max_length",
            max_length=tokenizer.model_max_length,
            truncation=True,
            return_tensors="pt",
        )
        text_input_ids = text_inputs.input_ids
        prompt_embeds = text_encoder(text_input_ids.to(device), attention_mask=None)[0]
        embeddings.append(prompt_embeds)
    return torch.concatenate(embeddings, dim=0).mean(dim=0).unsqueeze(0)

source_embeds = embed_prompts(source_prompts, pipeline.tokenizer, pipeline.text_encoder)
target_embeds = embed_prompts(target_prompts, pipeline.tokenizer, pipeline.text_encoder)
```

마지막으로, 임베딩을 `generate_mask()` 및 `invert()` 함수와 파이프라인에 전달하여 이미지를 생성합니다:

```diff
  from diffusers import DDIMInverseScheduler, DDIMScheduler
  from diffusers.utils import load_image, make_image_grid
  from PIL import Image

  pipeline.scheduler = DDIMScheduler.from_config(pipeline.scheduler.config)
  pipeline.inverse_scheduler = DDIMInverseScheduler.from_config(pipeline.scheduler.config)

  img_url = "https://github.com/Xiang-cd/DiffEdit-stable-diffusion/raw/main/assets/origin.png"
  raw_image = load_image(img_url).resize((768, 768))

  mask_image = pipeline.generate_mask(
      image=raw_image,
-     source_prompt=source_prompt,
-     target_prompt=target_prompt,
+     source_prompt_embeds=source_embeds,
+     target_prompt_embeds=target_embeds,
  )

  inv_latents = pipeline.invert(
-     prompt=source_prompt,
+     prompt_embeds=source_embeds,
      image=raw_image,
  ).latents

  output_image = pipeline(
      mask_image=mask_image,
      image_latents=inv_latents,
-     prompt=target_prompt,
-     negative_prompt=source_prompt,
+     prompt_embeds=target_embeds,
+     negative_prompt_embeds=source_embeds,
  ).images[0]
  mask_image = Image.fromarray((mask_image.squeeze()*255).astype("uint8"), "L")
  make_image_grid([raw_image, mask_image, output_image], rows=1, cols=3)
```

## 반전을 위한 캡션 생성하기

`source_prompt`를 캡션으로 사용하여 부분적으로 반전된 latents를 생성할 수 있지만, [BLIP](https://huggingface.co/docs/transformers/model_doc/blip) 모델을 사용하여 캡션을 자동으로 생성할 수도 있습니다.

🤗 Transformers 라이브러리에서 BLIP 모델과 프로세서를 불러옵니다:

```py
import torch
from transformers import BlipForConditionalGeneration, BlipProcessor

processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base", torch_dtype=torch.float16, low_cpu_mem_usage=True)
```

입력 이미지에서 캡션을 생성하는 유틸리티 함수를 만듭니다:

```py
@torch.no_grad()
def generate_caption(images, caption_generator, caption_processor):
    text = "a photograph of"

    inputs = caption_processor(images, text, return_tensors="pt").to(device="cuda", dtype=caption_generator.dtype)
    caption_generator.to("cuda")
    outputs = caption_generator.generate(**inputs, max_new_tokens=128)

    # 캡션 generator 오프로드
    caption_generator.to("cpu")

    caption = caption_processor.batch_decode(outputs, skip_special_tokens=True)[0]
    return caption
```

입력 이미지를 불러오고 `generate_caption` 함수를 사용하여 해당 이미지에 대한 캡션을 생성합니다:

```py
from diffusers.utils import load_image

img_url = "https://github.com/Xiang-cd/DiffEdit-stable-diffusion/raw/main/assets/origin.png"
raw_image = load_image(img_url).resize((768, 768))
caption = generate_caption(raw_image, model, processor)
```

<div class="flex justify-center">
    <figure>
        <img class="rounded-xl" src="https://github.com/Xiang-cd/DiffEdit-stable-diffusion/raw/main/assets/origin.png"/>
        <figcaption class="text-center">generated caption: "a photograph of a bowl of fruit on a table"</figcaption>
    </figure>
</div>

이제 캡션을 `invert()` 함수에 놓아 부분적으로 반전된 latents를 생성할 수 있습니다!


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/diffedit.md" />

### 어댑터 불러오기
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/loading_adapters.md

# 어댑터 불러오기


특정 물체의 이미지 또는 특정 스타일의 이미지를 생성하도록 diffusion 모델을 개인화하기 위한 몇 가지 [학습](../training/overview) 기법이 있습니다. 이러한 학습 방법은 각각 다른 유형의 어댑터를 생성합니다. 일부 어댑터는 완전히 새로운 모델을 생성하는 반면, 다른 어댑터는 임베딩 또는 가중치의 작은 부분만 수정합니다. 이는 각 어댑터의 로딩 프로세스도 다르다는 것을 의미합니다.

이 가이드에서는 DreamBooth, textual inversion 및 LoRA 가중치를 불러오는 방법을 설명합니다.

> [!TIP]
> 사용할 체크포인트와 임베딩은 [Stable Diffusion Conceptualizer](https://huggingface.co/spaces/sd-concepts-library/stable-diffusion-conceptualizer), [LoRA the Explorer](https://huggingface.co/spaces/multimodalart/LoraTheExplorer), [Diffusers Models Gallery](https://huggingface.co/spaces/huggingface-projects/diffusers-gallery)에서 찾아보시기 바랍니다.

## DreamBooth

[DreamBooth](https://dreambooth.github.io/)는 물체의 여러 이미지에 대한 *diffusion 모델 전체*를 미세 조정하여 새로운 스타일과 설정으로 해당 물체의 이미지를 생성합니다. 이 방법은 모델이 물체 이미지와 연관시키는 방법을 학습하는 프롬프트에 특수 단어를 사용하는 방식으로 작동합니다. 모든 학습 방법 중에서 드림부스는 전체 체크포인트 모델이기 때문에 파일 크기가 가장 큽니다(보통 몇 GB).

Hergé가 그린 단 10개의 이미지로 학습된 [herge_style](https://huggingface.co/sd-dreambooth-library/herge-style) 체크포인트를 불러와 해당 스타일의 이미지를 생성해 보겠습니다. 이 모델이 작동하려면 체크포인트를 트리거하는 프롬프트에 특수 단어 `herge_style`을 포함시켜야 합니다:

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("sd-dreambooth-library/herge-style", torch_dtype=torch.float16).to("cuda")
prompt = "A cute herge_style brown bear eating a slice of pizza, stunning color scheme, masterpiece, illustration"
image = pipeline(prompt).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/load_dreambooth.png" />
</div>

## Textual inversion

[Textual inversion](https://textual-inversion.github.io/)은 DreamBooth와 매우 유사하며 몇 개의 이미지만으로 특정 개념(스타일, 개체)을 생성하는 diffusion 모델을 개인화할 수도 있습니다. 이 방법은 프롬프트에 특정 단어를 입력하면 해당 이미지를 나타내는 새로운 임베딩을 학습하고 찾아내는 방식으로 작동합니다. 결과적으로 diffusion 모델 가중치는 동일하게 유지되고 훈련 프로세스는 비교적 작은(수 KB) 파일을 생성합니다.

Textual inversion은 임베딩을 생성하기 때문에 DreamBooth처럼 단독으로 사용할 수 없으며 또 다른 모델이 필요합니다.

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16).to("cuda")
```

이제 `load_textual_inversion()` 메서드를 사용하여 textual inversion 임베딩을 불러와 이미지를 생성할 수 있습니다. [sd-concepts-library/gta5-artwork](https://huggingface.co/sd-concepts-library/gta5-artwork) 임베딩을 불러와 보겠습니다. 이를 트리거하려면 프롬프트에 특수 단어 `<gta5-artwork>`를 포함시켜야 합니다:

```py
pipeline.load_textual_inversion("sd-concepts-library/gta5-artwork")
prompt = "A cute brown bear eating a slice of pizza, stunning color scheme, masterpiece, illustration, <gta5-artwork> style"
image = pipeline(prompt).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/load_txt_embed.png" />
</div>

Textual inversion은 또한 바람직하지 않은 사물에 대해 *네거티브 임베딩*을 생성하여 모델이 흐릿한 이미지나 손의 추가 손가락과 같은 바람직하지 않은 사물이 포함된 이미지를 생성하지 못하도록 학습할 수도 있습니다. 이는 프롬프트를 빠르게 개선하는 것이 쉬운 방법이 될 수 있습니다. 이는 이전과 같이 임베딩을 `load_textual_inversion()`으로 불러오지만 이번에는 두 개의 매개변수가 더 필요합니다:

- `weight_name`: 파일이 특정 이름의 🤗 Diffusers 형식으로 저장된 경우이거나 파일이 A1111 형식으로 저장된 경우, 불러올 가중치 파일을 지정합니다.
- `token`: 임베딩을 트리거하기 위해 프롬프트에서 사용할 특수 단어를 지정합니다.

[sayakpaul/EasyNegative-test](https://huggingface.co/sayakpaul/EasyNegative-test) 임베딩을 불러와 보겠습니다:

```py
pipeline.load_textual_inversion(
    "sayakpaul/EasyNegative-test", weight_name="EasyNegative.safetensors", token="EasyNegative"
)
```

이제 `token`을 사용해 네거티브 임베딩이 있는 이미지를 생성할 수 있습니다:

```py
prompt = "A cute brown bear eating a slice of pizza, stunning color scheme, masterpiece, illustration, EasyNegative"
negative_prompt = "EasyNegative"

image = pipeline(prompt, negative_prompt=negative_prompt, num_inference_steps=50).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/load_neg_embed.png" />
</div>

## LoRA

[Low-Rank Adaptation (LoRA)](https://huggingface.co/papers/2106.09685)은 속도가 빠르고 파일 크기가 (수백 MB로) 작기 때문에 널리 사용되는 학습 기법입니다. 이 가이드의 다른 방법과 마찬가지로, LoRA는 몇 장의 이미지만으로 새로운 스타일을 학습하도록 모델을 학습시킬 수 있습니다. 이는 diffusion 모델에 새로운 가중치를 삽입한 다음 전체 모델 대신 새로운 가중치만 학습시키는 방식으로 작동합니다. 따라서 LoRA를 더 빠르게 학습시키고 더 쉽게 저장할 수 있습니다.

> [!TIP]
> LoRA는 다른 학습 방법과 함께 사용할 수 있는 매우 일반적인 학습 기법입니다. 예를 들어, DreamBooth와 LoRA로 모델을 학습하는 것이 일반적입니다. 또한 새롭고 고유한 이미지를 생성하기 위해 여러 개의 LoRA를 불러오고 병합하는 것이 점점 더 일반화되고 있습니다. 병합은 이 불러오기 가이드의 범위를 벗어나므로 자세한 내용은 심층적인 [LoRA 병합](merge_loras) 가이드에서 확인할 수 있습니다.

LoRA는 다른 모델과 함께 사용해야 합니다:

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
```

그리고 `load_lora_weights()` 메서드를 사용하여 [ostris/super-cereal-sdxl-lora](https://huggingface.co/ostris/super-cereal-sdxl-lora) 가중치를 불러오고 리포지토리에서 가중치 파일명을 지정합니다:

```py
pipeline.load_lora_weights("ostris/super-cereal-sdxl-lora", weight_name="cereal_box_sdxl_v1.safetensors")
prompt = "bears, pizza bites"
image = pipeline(prompt).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/load_lora.png" />
</div>

`load_lora_weights()` 메서드는 LoRA 가중치를 UNet과 텍스트 인코더에 모두 불러옵니다. 이 메서드는 해당 케이스에서 LoRA를 불러오는 데 선호되는 방식입니다:

- LoRA 가중치에 UNet 및 텍스트 인코더에 대한 별도의 식별자가 없는 경우
- LoRA 가중치에 UNet과 텍스트 인코더에 대한 별도의 식별자가 있는 경우

하지만 LoRA 가중치만 UNet에 로드해야 하는 경우에는 `load_attn_procs()` 메서드를 사용할 수 있습니다. [jbilcke-hf/sdxl-cinematic-1](https://huggingface.co/jbilcke-hf/sdxl-cinematic-1) LoRA를 불러와 보겠습니다:

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
pipeline.unet.load_attn_procs("jbilcke-hf/sdxl-cinematic-1", weight_name="pytorch_lora_weights.safetensors")

# 프롬프트에서 cnmt를 사용하여 LoRA를 트리거합니다.
prompt = "A cute cnmt eating a slice of pizza, stunning color scheme, masterpiece, illustration"
image = pipeline(prompt).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/load_attn_proc.png" />
</div>

LoRA 가중치를 언로드하려면 `unload_lora_weights()` 메서드를 사용하여 LoRA 가중치를 삭제하고 모델을 원래 가중치로 복원합니다:

```py
pipeline.unload_lora_weights()
```

### LoRA 가중치 스케일 조정하기

`load_lora_weights()` 및 `load_attn_procs()` 모두 `cross_attention_kwargs={"scale": 0.5}` 파라미터를 전달하여 얼마나 LoRA 가중치를 사용할지 조정할 수 있습니다. 값이 `0`이면 기본 모델 가중치만 사용하는 것과 같고, 값이 `1`이면 완전히 미세 조정된 LoRA를 사용하는 것과 같습니다.

레이어당 사용되는 LoRA 가중치의 양을 보다 세밀하게 제어하려면 `set_adapters()`를 사용하여 각 레이어의 가중치를 얼마만큼 조정할지 지정하는 딕셔너리를 전달할 수 있습니다.
```python
pipe = ... # 파이프라인 생성
pipe.load_lora_weights(..., adapter_name="my_adapter")
scales = {
    "text_encoder": 0.5,
    "text_encoder_2": 0.5,  # 파이프에 두 번째 텍스트 인코더가 있는 경우에만 사용 가능
    "unet": {
        "down": 0.9,  # down 부분의 모든 트랜스포머는 스케일 0.9를 사용
        # "mid"  # 이 예제에서는 "mid"가 지정되지 않았으므로 중간 부분의 모든 트랜스포머는 기본 스케일 1.0을 사용
        "up": {
            "block_0": 0.6,  # # up의 0번째 블록에 있는 3개의 트랜스포머는 모두 스케일 0.6을 사용
            "block_1": [0.4, 0.8, 1.0],  # up의 첫 번째 블록에 있는 3개의 트랜스포머는 각각 스케일 0.4, 0.8, 1.0을 사용
        }
    }
}
pipe.set_adapters("my_adapter", scales)
```

이는 여러 어댑터에서도 작동합니다. 방법은 [이 가이드](https://huggingface.co/docs/diffusers/tutorials/using_peft_for_inference#customize-adapters-strength)를 참조하세요.

> [!WARNING]
> 현재 `set_adapters()`는 어텐션 가중치의 스케일링만 지원합니다. LoRA에 다른 부분(예: resnets or down-/upsamplers)이 있는 경우 1.0의 스케일을 유지합니다.

### Kohya와 TheLastBen

커뮤니티에서 인기 있는 다른 LoRA trainer로는 [Kohya](https://github.com/kohya-ss/sd-scripts/)와 [TheLastBen](https://github.com/TheLastBen/fast-stable-diffusion)의 trainer가 있습니다. 이 trainer들은 🤗 Diffusers가 훈련한 것과는 다른 LoRA 체크포인트를 생성하지만, 같은 방식으로 불러올 수 있습니다.

<hfoptions id="other-trainers">
<hfoption id="Kohya">

Kohya LoRA를 불러오기 위해, 예시로 [Civitai](https://civitai.com/)에서 [Blueprintify SD XL 1.0](https://civitai.com/models/150986/blueprintify-sd-xl-10) 체크포인트를 다운로드합니다:

```sh
!wget https://civitai.com/api/download/models/168776 -O blueprintify-sd-xl-10.safetensors
```

LoRA 체크포인트를 `load_lora_weights()` 메서드로 불러오고 `weight_name` 파라미터에 파일명을 지정합니다:

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
pipeline.load_lora_weights("path/to/weights", weight_name="blueprintify-sd-xl-10.safetensors")
```

이미지를 생성합니다:

```py
# LoRA를 트리거하기 위해 bl3uprint를 프롬프트에 사용
prompt = "bl3uprint, a highly detailed blueprint of the eiffel tower, explaining how to build all parts, many txt, blueprint grid backdrop"
image = pipeline(prompt).images[0]
image
```

> [!WARNING]
> Kohya LoRA를 🤗 Diffusers와 함께 사용할 때 몇 가지 제한 사항이 있습니다:
>
> - [여기](https://github.com/huggingface/diffusers/pull/4287/#issuecomment-1655110736)에 설명된 여러 가지 이유로 인해 이미지가 ComfyUI와 같은 UI에서 생성된 이미지와 다르게 보일 수 있습니다.
> - [LyCORIS 체크포인트](https://github.com/KohakuBlueleaf/LyCORIS)가 완전히 지원되지 않습니다. `load_lora_weights()` 메서드는 LoRA 및 LoCon 모듈로 LyCORIS 체크포인트를 불러올 수 있지만, Hada 및 LoKR은 지원되지 않습니다.

</hfoption>
<hfoption id="TheLastBen">

TheLastBen에서 체크포인트를 불러오는 방법은 매우 유사합니다. 예를 들어, [TheLastBen/William_Eggleston_Style_SDXL](https://huggingface.co/TheLastBen/William_Eggleston_Style_SDXL) 체크포인트를 불러오려면:

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16).to("cuda")
pipeline.load_lora_weights("TheLastBen/William_Eggleston_Style_SDXL", weight_name="wegg.safetensors")

# LoRA를 트리거하기 위해 william eggleston를 프롬프트에 사용
prompt = "a house by william eggleston, sunrays, beautiful, sunlight, sunrays, beautiful"
image = pipeline(prompt=prompt).images[0]
image
```

</hfoption>
</hfoptions>

## IP-Adapter

[IP-Adapter](https://ip-adapter.github.io/)는 모든 diffusion 모델에 이미지 프롬프트를 사용할 수 있는 경량 어댑터입니다. 이 어댑터는 이미지와 텍스트 feature의 cross-attention 레이어를 분리하여 작동합니다. 다른 모든 모델 컴포넌트튼 freeze되고 UNet의 embedded 이미지 features만 학습됩니다. 따라서 IP-Adapter 파일은 일반적으로 최대 100MB에 불과합니다.

다양한 작업과 구체적인 사용 사례에 IP-Adapter를 사용하는 방법에 대한 자세한 내용은 [IP-Adapter](../using-diffusers/ip_adapter) 가이드에서 확인할 수 있습니다.

> [!TIP]
> Diffusers는 현재 가장 많이 사용되는 일부 파이프라인에 대해서만 IP-Adapter를 지원합니다. 멋진 사용 사례가 있는 지원되지 않는 파이프라인에 IP-Adapter를 통합하고 싶다면 언제든지 기능 요청을 여세요!
> 공식 IP-Adapter 체크포인트는 [h94/IP-Adapter](https://huggingface.co/h94/IP-Adapter)에서 확인할 수 있습니다.

시작하려면 Stable Diffusion 체크포인트를 불러오세요.

```py
from diffusers import AutoPipelineForText2Image
import torch
from diffusers.utils import load_image

pipeline = AutoPipelineForText2Image.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16).to("cuda")
```

그런 다음 IP-Adapter 가중치를 불러와 `load_ip_adapter()` 메서드를 사용하여 파이프라인에 추가합니다.

```py
pipeline.load_ip_adapter("h94/IP-Adapter", subfolder="models", weight_name="ip-adapter_sd15.bin")
```

불러온 뒤, 이미지 및 텍스트 프롬프트가 있는 파이프라인을 사용하여 이미지 생성 프로세스를 가이드할 수 있습니다.

```py
image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/load_neg_embed.png")
generator = torch.Generator(device="cpu").manual_seed(33)
images = pipeline(
    prompt='best quality, high quality, wearing sunglasses',
    ip_adapter_image=image,
    negative_prompt="monochrome, lowres, bad anatomy, worst quality, low quality",
    num_inference_steps=50,
    generator=generator,
).images[0]
images
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/YiYiXu/testing-images/resolve/main/ip-bear.png" />
</div>

### IP-Adapter Plus

IP-Adapter는 이미지 인코더를 사용하여 이미지 feature를 생성합니다. IP-Adapter 리포지토리에 `image_encoder` 하위 폴더가 있는 경우, 이미지 인코더가 자동으로 불러와 파이프라인에 등록됩니다. 그렇지 않은 경우, `CLIPVisionModelWithProjection` 모델을 사용하여 이미지 인코더를 명시적으로 불러와 파이프라인에 전달해야 합니다.

이는 ViT-H 이미지 인코더를 사용하는 *IP-Adapter Plus* 체크포인트에 해당하는 케이스입니다.

```py
from transformers import CLIPVisionModelWithProjection

image_encoder = CLIPVisionModelWithProjection.from_pretrained(
    "h94/IP-Adapter",
    subfolder="models/image_encoder",
    torch_dtype=torch.float16
)

pipeline = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    image_encoder=image_encoder,
    torch_dtype=torch.float16
).to("cuda")

pipeline.load_ip_adapter("h94/IP-Adapter", subfolder="sdxl_models", weight_name="ip-adapter-plus_sdxl_vit-h.safetensors")
```

### IP-Adapter Face ID 모델

IP-Adapter FaceID 모델은 CLIP 이미지 임베딩 대신 `insightface`에서 생성한 이미지 임베딩을 사용하는 실험적인 IP Adapter입니다. 이러한 모델 중 일부는 LoRA를 사용하여 ID 일관성을 개선하기도 합니다.
이러한 모델을 사용하려면 `insightface`와 해당 요구 사항을 모두 설치해야 합니다.

> [!WARNING]
> InsightFace 사전학습된 모델은 비상업적 연구 목적으로만 사용할 수 있으므로, IP-Adapter-FaceID 모델은 연구 목적으로만 릴리즈되었으며 상업적 용도로는 사용할 수 없습니다.

```py
pipeline = AutoPipelineForText2Image.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    torch_dtype=torch.float16
).to("cuda")

pipeline.load_ip_adapter("h94/IP-Adapter-FaceID", subfolder=None, weight_name="ip-adapter-faceid_sdxl.bin", image_encoder_folder=None)
```

두 가지 IP 어댑터 FaceID Plus 모델 중 하나를 사용하려는 경우, 이 모델들은 더 나은 사실감을 얻기 위해 `insightface`와 CLIP 이미지 임베딩을 모두 사용하므로, CLIP 이미지 인코더도 불러와야 합니다.

```py
from transformers import CLIPVisionModelWithProjection

image_encoder = CLIPVisionModelWithProjection.from_pretrained(
    "laion/CLIP-ViT-H-14-laion2B-s32B-b79K",
    torch_dtype=torch.float16,
)

pipeline = AutoPipelineForText2Image.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    image_encoder=image_encoder,
    torch_dtype=torch.float16
).to("cuda")

pipeline.load_ip_adapter("h94/IP-Adapter-FaceID", subfolder=None, weight_name="ip-adapter-faceid-plus_sd15.bin")
```


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/loading_adapters.md" />

### 다양한 Stable Diffusion 포맷 불러오기
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/other-formats.md

# 다양한 Stable Diffusion 포맷 불러오기

Stable Diffusion 모델들은 학습 및 저장된 프레임워크와 다운로드 위치에 따라 다양한 형식으로 제공됩니다. 이러한 형식을 🤗 Diffusers에서 사용할 수 있도록 변환하면 추론을 위한 [다양한 스케줄러 사용](schedulers), 사용자 지정 파이프라인 구축, 추론 속도 최적화를 위한 다양한 기법과 방법 등 라이브러리에서 지원하는 모든 기능을 사용할 수 있습니다.

> [!TIP]
> 우리는 `.safetensors` 형식을 추천합니다. 왜냐하면 기존의 pickled 파일은 취약하고 머신에서 코드를 실행할 때 악용될 수 있는 것에 비해 훨씬 더 안전합니다. (safetensors 불러오기 가이드에서 자세히 알아보세요.)

이 가이드에서는 다른 Stable Diffusion 형식을 🤗 Diffusers와 호환되도록 변환하는 방법을 설명합니다.

## PyTorch .ckpt

체크포인트 또는 `.ckpt` 형식은 일반적으로 모델을 저장하는 데 사용됩니다. `.ckpt` 파일은 전체 모델을 포함하며 일반적으로 크기가 몇 GB입니다. `.ckpt` 파일을 [~StableDiffusionPipeline.from_ckpt] 메서드를 사용하여 직접 불러와서 사용할 수도 있지만, 일반적으로 두 가지 형식을 모두 사용할 수 있도록 `.ckpt` 파일을 🤗 Diffusers로 변환하는 것이 더 좋습니다.

`.ckpt` 파일을 변환하는 두 가지 옵션이 있습니다. Space를 사용하여 체크포인트를 변환하거나 스크립트를 사용하여 `.ckpt` 파일을 변환합니다.

### Space로 변환하기

`.ckpt` 파일을 변환하는 가장 쉽고 편리한 방법은 SD에서 Diffusers로 스페이스를 사용하는 것입니다. Space의 지침에 따라 .ckpt 파일을 변환 할 수 있습니다.

이 접근 방식은 기본 모델에서는 잘 작동하지만 더 많은 사용자 정의 모델에서는 어려움을 겪을 수 있습니다. 빈 pull request나 오류를 반환하면 Space가 실패한 것입니다.
이 경우 스크립트를 사용하여 `.ckpt` 파일을 변환해 볼 수 있습니다.

### 스크립트로 변환하기

🤗 Diffusers는 `.ckpt`  파일 변환을 위한 변환 스크립트를 제공합니다. 이 접근 방식은 위의 Space보다 더 안정적입니다.

시작하기 전에 스크립트를 실행할 🤗 Diffusers의 로컬 클론(clone)이 있는지 확인하고 Hugging Face 계정에 로그인하여 pull request를 열고 변환된 모델을 허브에 푸시할 수 있도록 하세요.

```bash
hf auth login
```

스크립트를 사용하려면:

1. 변환하려는 `.ckpt`  파일이 포함된 리포지토리를 Git으로 클론(clone)합니다.

이 예제에서는 TemporalNet .ckpt 파일을 변환해 보겠습니다:

```bash
git lfs install
git clone https://huggingface.co/CiaraRowles/TemporalNet
```

2. 체크포인트를 변환할 리포지토리에서 pull request를 엽니다:

```bash
cd TemporalNet && git fetch origin refs/pr/13:pr/13
git checkout pr/13
```

3. 변환 스크립트에서 구성할 입력 인수는 여러 가지가 있지만 가장 중요한 인수는 다음과 같습니다:

- `checkpoint_path`: 변환할 `.ckpt` 파일의 경로를 입력합니다.
- `original_config_file`: 원래 아키텍처의 구성을 정의하는 YAML 파일입니다. 이 파일을 찾을 수 없는 경우 `.ckpt` 파일을 찾은 GitHub 리포지토리에서 YAML 파일을 검색해 보세요.
- `dump_path`: 변환된 모델의 경로

예를 들어, TemporalNet 모델은 Stable Diffusion v1.5 및 ControlNet 모델이기 때문에 ControlNet 리포지토리에서 cldm_v15.yaml 파일을 가져올 수 있습니다.

4. 이제 스크립트를 실행하여 .ckpt 파일을 변환할 수 있습니다:

```bash
python ../diffusers/scripts/convert_original_stable_diffusion_to_diffusers.py --checkpoint_path temporalnetv3.ckpt --original_config_file cldm_v15.yaml --dump_path ./ --controlnet
```

5. 변환이 완료되면 변환된 모델을 업로드하고 결과물을 pull request [pull request](https://huggingface.co/CiaraRowles/TemporalNet/discussions/13)를 테스트하세요!

```bash
git push origin pr/13:refs/pr/13
```

## **Keras .pb or .h5**

🧪 이 기능은 실험적인 기능입니다. 현재로서는 Stable Diffusion v1 체크포인트만 변환 KerasCV Space에서 지원됩니다.

[KerasCV](https://keras.io/keras_cv/)는 [Stable Diffusion](https://github.com/keras-team/keras-cv/blob/master/keras_cv/models/stable_diffusion)  v1 및 v2에 대한 학습을 지원합니다. 그러나 추론 및 배포를 위한 Stable Diffusion 모델 실험을 제한적으로 지원하는 반면, 🤗 Diffusers는 다양한 [noise schedulers](https://huggingface.co/docs/diffusers/using-diffusers/schedulers), [flash attention](https://huggingface.co/docs/diffusers/optimization/xformers), and [other optimization techniques](https://huggingface.co/docs/diffusers/optimization/fp16) 등 이러한 목적을 위한 보다 완벽한 기능을 갖추고 있습니다.

[Convert KerasCV](https://huggingface.co/spaces/sayakpaul/convert-kerascv-sd-diffusers) Space 변환은 `.pb` 또는 `.h5`을 PyTorch로 변환한 다음, 추론할 수 있도록 `StableDiffusionPipeline` 으로 감싸서 준비합니다. 변환된 체크포인트는 Hugging Face Hub의 리포지토리에 저장됩니다.

예제로, textual-inversion으로 학습된 `[sayakpaul/textual-inversion-kerasio](https://huggingface.co/sayakpaul/textual-inversion-kerasio/tree/main)` 체크포인트를 변환해 보겠습니다. 이것은 특수 토큰  `<my-funny-cat>`을 사용하여 고양이로 이미지를 개인화합니다.

KerasCV Space 변환에서는 다음을 입력할 수 있습니다:

- Hugging Face 토큰.
- UNet 과 텍스트 인코더(text encoder) 가중치를 다운로드하는 경로입니다. 모델을 어떻게 학습할지 방식에 따라, UNet과 텍스트 인코더의 경로를 모두 제공할 필요는 없습니다. 예를 들어, textual-inversion에는 텍스트 인코더의 임베딩만 필요하고 텍스트-이미지(text-to-image) 모델 변환에는 UNet 가중치만 필요합니다.
- Placeholder 토큰은 textual-inversion 모델에만 적용됩니다.
- `output_repo_prefix`는 변환된 모델이 저장되는 리포지토리의 이름입니다.

**Submit** (제출) 버튼을 클릭하면 KerasCV 체크포인트가 자동으로 변환됩니다! 체크포인트가 성공적으로 변환되면, 변환된 체크포인트가 포함된 새 리포지토리로 연결되는 링크가 표시됩니다. 새 리포지토리로 연결되는 링크를 따라가면 변환된 모델을 사용해 볼 수 있는 추론 위젯이 포함된 모델 카드가 생성된 KerasCV Space 변환을 확인할 수 있습니다.

코드를 사용하여 추론을 실행하려면 모델 카드의 오른쪽 상단 모서리에 있는 **Use in Diffusers**  버튼을 클릭하여 예시 코드를 복사하여 붙여넣습니다:

```py
from diffusers import DiffusionPipeline

pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
```

그러면 다음과 같은 이미지를 생성할 수 있습니다:

```py
from diffusers import DiffusionPipeline

pipeline = DiffusionPipeline.from_pretrained("sayakpaul/textual-inversion-cat-kerascv_sd_diffusers_pipeline")
pipeline.to("cuda")

placeholder_token = "<my-funny-cat-token>"
prompt = f"two {placeholder_token} getting married, photorealistic, high quality"
image = pipeline(prompt, num_inference_steps=50).images[0]
```

## **A1111 LoRA files**

[Automatic1111](https://github.com/AUTOMATIC1111/stable-diffusion-webui) (A1111)은 Stable Diffusion을 위해 널리 사용되는 웹 UI로, [Civitai](https://civitai.com/) 와 같은 모델 공유 플랫폼을 지원합니다. 특히 LoRA 기법으로 학습된 모델은 학습 속도가 빠르고 완전히 파인튜닝된 모델보다 파일 크기가 훨씬 작기 때문에 인기가 높습니다.

🤗 Diffusers는 `load_lora_weights()`:를 사용하여 A1111 LoRA 체크포인트 불러오기를 지원합니다:

```py
from diffusers import DiffusionPipeline, UniPCMultistepScheduler
import torch

pipeline = DiffusionPipeline.from_pretrained(
    "andite/anything-v4.0", torch_dtype=torch.float16, safety_checker=None
).to("cuda")
pipeline.scheduler = UniPCMultistepScheduler.from_config(pipeline.scheduler.config)
```

Civitai에서 LoRA 체크포인트를 다운로드하세요; 이 예제에서는  [Howls Moving Castle,Interior/Scenery LoRA (Ghibli Stlye)](https://civitai.com/models/14605?modelVersionId=19998) 체크포인트를 사용했지만, 어떤 LoRA 체크포인트든 자유롭게 사용해 보세요!

```bash
!wget https://civitai.com/api/download/models/19998 -O howls_moving_castle.safetensors
```

메서드를 사용하여 파이프라인에 LoRA 체크포인트를 불러옵니다:

```py
pipeline.load_lora_weights(".", weight_name="howls_moving_castle.safetensors")
```

이제 파이프라인을 사용하여 이미지를 생성할 수 있습니다:

```py
prompt = "masterpiece, illustration, ultra-detailed, cityscape, san francisco, golden gate bridge, california, bay area, in the snow, beautiful detailed starry sky"
negative_prompt = "lowres, cropped, worst quality, low quality, normal quality, artifacts, signature, watermark, username, blurry, more than one bridge, bad architecture"

images = pipeline(
    prompt=prompt,
    negative_prompt=negative_prompt,
    width=512,
    height=512,
    num_inference_steps=25,
    num_images_per_prompt=4,
    generator=torch.manual_seed(0),
).images
```

마지막으로, 디스플레이에 이미지를 표시하는 헬퍼 함수를 만듭니다:

```py
from PIL import Image


def image_grid(imgs, rows=2, cols=2):
    w, h = imgs[0].size
    grid = Image.new("RGB", size=(cols * w, rows * h))

    for i, img in enumerate(imgs):
        grid.paste(img, box=(i % cols * w, i // cols * h))
    return grid


image_grid(images)
```

<div class="flex justify-center">
  <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/a1111-lora-sf.png" />
</div>


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/other-formats.md" />

### 파일들을 Hub로 푸시하기
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/push_to_hub.md

# 파일들을 Hub로 푸시하기


🤗 Diffusers는 모델, 스케줄러 또는 파이프라인을 Hub에 업로드할 수 있는 `PushToHubMixin`을 제공합니다. 이는 Hub에 당신의 파일을 저장하는 쉬운 방법이며, 다른 사람들과 작업을 공유할 수도 있습니다. 실제적으로 `PushToHubMixin`가 동작하는 방식은 다음과 같습니다:

1. Hub에 리포지토리를 생성합니다.
2. 나중에 다시 불러올 수 있도록 모델, 스케줄러 또는 파이프라인 파일을 저장합니다.
3. 이러한 파일이 포함된 폴더를 Hub에 업로드합니다.

이 가이드는 `PushToHubMixin`을 사용하여 Hub에 파일을 업로드하는 방법을 보여줍니다.

먼저 액세스 [토큰](https://huggingface.co/settings/tokens)으로 Hub 계정에 로그인해야 합니다:

```py
from huggingface_hub import notebook_login

notebook_login()
```

## 모델

모델을 허브에 푸시하려면 `push_to_hub()`를 호출하고 Hub에 저장할 모델의 리포지토리 id를 지정합니다:

```py
from diffusers import ControlNetModel

controlnet = ControlNetModel(
    block_out_channels=(32, 64),
    layers_per_block=2,
    in_channels=4,
    down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
    cross_attention_dim=32,
    conditioning_embedding_out_channels=(16, 32),
)
controlnet.push_to_hub("my-controlnet-model")
```

모델의 경우 Hub에 푸시할 가중치의 [*변형*](loading#checkpoint-variants)을 지정할 수도 있습니다. 예를 들어, `fp16` 가중치를 푸시하려면 다음과 같이 하세요:

```py
controlnet.push_to_hub("my-controlnet-model", variant="fp16")
```

`push_to_hub()` 함수는 모델의 `config.json` 파일을 저장하고 가중치는 `safetensors` 형식으로 자동으로 저장됩니다.

이제 Hub의 리포지토리에서 모델을 다시 불러올 수 있습니다:

```py
model = ControlNetModel.from_pretrained("your-namespace/my-controlnet-model")
```

## 스케줄러

스케줄러를 허브에 푸시하려면 `push_to_hub()`를 호출하고 Hub에 저장할 스케줄러의 리포지토리 id를 지정합니다:

```py
from diffusers import DDIMScheduler

scheduler = DDIMScheduler(
    beta_start=0.00085,
    beta_end=0.012,
    beta_schedule="scaled_linear",
    clip_sample=False,
    set_alpha_to_one=False,
)
scheduler.push_to_hub("my-controlnet-scheduler")
```

`push_to_hub()` 함수는 스케줄러의 `scheduler_config.json` 파일을 지정된 리포지토리에 저장합니다.

이제 허브의 리포지토리에서 스케줄러를 다시 불러올 수 있습니다:

```py
scheduler = DDIMScheduler.from_pretrained("your-namepsace/my-controlnet-scheduler")
```

## 파이프라인

모든 컴포넌트가 포함된 전체 파이프라인을 Hub로 푸시할 수도 있습니다. 예를 들어, 원하는 파라미터로 `StableDiffusionPipeline`의 컴포넌트들을 초기화합니다:

```py
from diffusers import (
    UNet2DConditionModel,
    AutoencoderKL,
    DDIMScheduler,
    StableDiffusionPipeline,
)
from transformers import CLIPTextModel, CLIPTextConfig, CLIPTokenizer

unet = UNet2DConditionModel(
    block_out_channels=(32, 64),
    layers_per_block=2,
    sample_size=32,
    in_channels=4,
    out_channels=4,
    down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
    up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
    cross_attention_dim=32,
)

scheduler = DDIMScheduler(
    beta_start=0.00085,
    beta_end=0.012,
    beta_schedule="scaled_linear",
    clip_sample=False,
    set_alpha_to_one=False,
)

vae = AutoencoderKL(
    block_out_channels=[32, 64],
    in_channels=3,
    out_channels=3,
    down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
    up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
    latent_channels=4,
)

text_encoder_config = CLIPTextConfig(
    bos_token_id=0,
    eos_token_id=2,
    hidden_size=32,
    intermediate_size=37,
    layer_norm_eps=1e-05,
    num_attention_heads=4,
    num_hidden_layers=5,
    pad_token_id=1,
    vocab_size=1000,
)
text_encoder = CLIPTextModel(text_encoder_config)
tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
```

모든 컴포넌트들을 `StableDiffusionPipeline`에 전달하고 `push_to_hub()`를 호출하여 파이프라인을 Hub로 푸시합니다:

```py
components = {
    "unet": unet,
    "scheduler": scheduler,
    "vae": vae,
    "text_encoder": text_encoder,
    "tokenizer": tokenizer,
    "safety_checker": None,
    "feature_extractor": None,
}

pipeline = StableDiffusionPipeline(**components)
pipeline.push_to_hub("my-pipeline")
```

`push_to_hub()` 함수는 각 컴포넌트를 리포지토리의 하위 폴더에 저장합니다. 이제 Hub의 리포지토리에서 파이프라인을 다시 불러올 수 있습니다:

```py
pipeline = StableDiffusionPipeline.from_pretrained("your-namespace/my-pipeline")
```

## 비공개

모델, 스케줄러 또는 파이프라인 파일들을 비공개로 두려면 `push_to_hub()` 함수에서 `private=True`를 설정하세요:

```py
controlnet.push_to_hub("my-controlnet-model-private", private=True)
```

비공개 리포지토리는 본인만 볼 수 있으며 다른 사용자는 리포지토리를 복제할 수 없고 리포지토리가 검색 결과에 표시되지 않습니다. 사용자가 비공개 리포지토리의 URL을 가지고 있더라도 `404 - Sorry, we can't find the page you are looking for`라는 메시지가 표시됩니다. 비공개 리포지토리에서 모델을 로드하려면 [로그인](https://huggingface.co/docs/huggingface_hub/quick-start#login) 상태여야 합니다.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/push_to_hub.md" />

### 조건부 이미지 생성
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/conditional_image_generation.md

# 조건부 이미지 생성


조건부 이미지 생성을 사용하면 텍스트 프롬프트에서 이미지를 생성할 수 있습니다. 텍스트는 임베딩으로 변환되며, 임베딩은 노이즈에서 이미지를 생성하도록 모델을 조건화하는 데 사용됩니다.

`DiffusionPipeline`은 추론을 위해 사전 훈련된 diffusion 시스템을 사용하는 가장 쉬운 방법입니다.

먼저 `DiffusionPipeline`의 인스턴스를 생성하고 다운로드할 파이프라인 [체크포인트](https://huggingface.co/models?library=diffusers&sort=downloads)를 지정합니다.

이 가이드에서는 [잠재 Diffusion](https://huggingface.co/CompVis/ldm-text2im-large-256)과 함께 텍스트-이미지 생성에 `DiffusionPipeline`을 사용합니다:

```python
>>> from diffusers import DiffusionPipeline

>>> generator = DiffusionPipeline.from_pretrained("CompVis/ldm-text2im-large-256")
```

`DiffusionPipeline`은 모든 모델링, 토큰화, 스케줄링 구성 요소를 다운로드하고 캐시합니다.
이 모델은 약 14억 개의 파라미터로 구성되어 있기 때문에 GPU에서 실행할 것을 강력히 권장합니다.
PyTorch에서와 마찬가지로 생성기 객체를 GPU로 이동할 수 있습니다:

```python
>>> generator.to("cuda")
```

이제 텍스트 프롬프트에서 `생성기`를 사용할 수 있습니다:

```python
>>> image = generator("An image of a squirrel in Picasso style").images[0]
```

출력값은 기본적으로 [`PIL.Image`](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class) 객체로 래핑됩니다.

호출하여 이미지를 저장할 수 있습니다:

```python
>>> image.save("image_of_squirrel_painting.png")
```

아래 스페이스를 사용해보고 안내 배율 매개변수를 자유롭게 조정하여 이미지 품질에 어떤 영향을 미치는지 확인해 보세요!

<iframe
	src="https://stabilityai-stable-diffusion.hf.space"
	frameborder="0"
	width="850"
	height="500"
></iframe>

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/conditional_image_generation.md" />

### 텍스트 기반 image-to-image 생성
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/img2img.md

# 텍스트 기반 image-to-image 생성


`StableDiffusionImg2ImgPipeline`을 사용하면 텍스트 프롬프트와 시작 이미지를 전달하여 새 이미지 생성의 조건을 지정할 수 있습니다.

시작하기 전에 필요한 라이브러리가 모두 설치되어 있는지 확인하세요:

```bash
!pip install diffusers transformers ftfy accelerate
```

[`nitrosocke/Ghibli-Diffusion`](https://huggingface.co/nitrosocke/Ghibli-Diffusion)과 같은 사전학습된 stable diffusion 모델로 `StableDiffusionImg2ImgPipeline`을 생성하여 시작하세요.


```python
import torch
import requests
from PIL import Image
from io import BytesIO
from diffusers import StableDiffusionImg2ImgPipeline

device = "cuda"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained("nitrosocke/Ghibli-Diffusion", torch_dtype=torch.float16).to(
    device
)
```

초기 이미지를 다운로드하고 사전 처리하여 파이프라인에 전달할 수 있습니다:

```python
url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"

response = requests.get(url)
init_image = Image.open(BytesIO(response.content)).convert("RGB")
init_image.thumbnail((768, 768))
init_image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/image_2_image_using_diffusers_cell_8_output_0.jpeg"/>
</div>

> [!TIP]
> 💡 `strength`는 입력 이미지에 추가되는 노이즈의 양을 제어하는 0.0에서 1.0 사이의 값입니다. 1.0에 가까운 값은 다양한 변형을 허용하지만 입력 이미지와 의미적으로 일치하지 않는 이미지를 생성합니다.

프롬프트를 정의하고(지브리 스타일(Ghibli-style)에 맞게 조정된 이 체크포인트의 경우 프롬프트 앞에 `ghibli style` 토큰을 붙여야 합니다) 파이프라인을 실행합니다:

```python
prompt = "ghibli style, a fantasy landscape with castles"
generator = torch.Generator(device=device).manual_seed(1024)
image = pipe(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5, generator=generator).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/ghibli-castles.png"/>
</div>

다른 스케줄러로 실험하여 출력에 어떤 영향을 미치는지 확인할 수도 있습니다:

```python
from diffusers import LMSDiscreteScheduler

lms = LMSDiscreteScheduler.from_config(pipe.scheduler.config)
pipe.scheduler = lms
generator = torch.Generator(device=device).manual_seed(1024)
image = pipe(prompt=prompt, image=init_image, strength=0.75, guidance_scale=7.5, generator=generator).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/lms-ghibli.png"/>
</div>

아래 공백을 확인하고 `strength` 값을 다르게 설정하여 이미지를 생성해 보세요. `strength`를 낮게 설정하면 원본 이미지와 더 유사한 이미지가 생성되는 것을 확인할 수 있습니다.

자유롭게 스케줄러를 `LMSDiscreteScheduler`로 전환하여 출력에 어떤 영향을 미치는지 확인해 보세요.

<iframe
	src="https://stevhliu-ghibli-img2img.hf.space"
	frameborder="0"
	width="850"
	height="500"
></iframe>

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/img2img.md" />

### Stable Diffusion XL Turbo
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/sdxl_turbo.md

# Stable Diffusion XL Turbo


SDXL Turbo는 adversarial time-distilled(적대적 시간 전이) [Stable Diffusion XL](https://huggingface.co/papers/2307.01952)(SDXL) 모델로, 단 한 번의 스텝만으로 추론을 실행할 수 있습니다.

이 가이드에서는 text-to-image와 image-to-image를 위한 SDXL-Turbo를 사용하는 방법을 설명합니다.

시작하기 전에 다음 라이브러리가 설치되어 있는지 확인하세요:

```py
# Colab에서 필요한 라이브러리를 설치하기 위해 주석을 제외하세요
#!pip install -q diffusers transformers accelerate
```

## 모델 체크포인트 불러오기

모델 가중치는 Hub의 별도 하위 폴더 또는 로컬에 저장할 수 있으며, 이 경우 `from_pretrained()` 메서드를 사용해야 합니다:

```py
from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
import torch

pipeline = AutoPipelineForText2Image.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16")
pipeline = pipeline.to("cuda")
```

또한 `from_single_file()` 메서드를 사용하여 허브 또는 로컬에서 단일 파일 형식(`.ckpt` 또는 `.safetensors`)으로 저장된 모델 체크포인트를 불러올 수도 있습니다:

```py
from diffusers import StableDiffusionXLPipeline
import torch

pipeline = StableDiffusionXLPipeline.from_single_file(
    "https://huggingface.co/stabilityai/sdxl-turbo/blob/main/sd_xl_turbo_1.0_fp16.safetensors", torch_dtype=torch.float16)
pipeline = pipeline.to("cuda")
```

## Text-to-image

Text-to-image의 경우 텍스트 프롬프트를 전달합니다. 기본적으로 SDXL Turbo는 512x512 이미지를 생성하며, 이 해상도에서 최상의 결과를 제공합니다. `height` 및 `width` 매개 변수를 768x768 또는 1024x1024로 설정할 수 있지만 이 경우 품질 저하를 예상할 수 있습니다.

모델이 `guidance_scale` 없이 학습되었으므로 이를 0.0으로 설정해 비활성화해야 합니다. 단일 추론 스텝만으로도 고품질 이미지를 생성할 수 있습니다.
스텝 수를 2, 3 또는 4로 늘리면 이미지 품질이 향상됩니다.

```py
from diffusers import AutoPipelineForText2Image
import torch

pipeline_text2image = AutoPipelineForText2Image.from_pretrained("stabilityai/sdxl-turbo", torch_dtype=torch.float16, variant="fp16")
pipeline_text2image = pipeline_text2image.to("cuda")

prompt = "A cinematic shot of a baby racoon wearing an intricate italian priest robe."

image = pipeline_text2image(prompt=prompt, guidance_scale=0.0, num_inference_steps=1).images[0]
image
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/sdxl-turbo-text2img.png" alt="generated image of a racoon in a robe"/>
</div>

## Image-to-image

Image-to-image 생성의 경우 `num_inference_steps * strength`가 1보다 크거나 같은지 확인하세요.
Image-to-image 파이프라인은 아래 예제에서 `0.5 * 2.0 = 1` 스텝과 같이 `int(num_inference_steps * strength)` 스텝으로 실행됩니다.

```py
from diffusers import AutoPipelineForImage2Image
from diffusers.utils import load_image, make_image_grid

# 체크포인트를 불러올 때 추가 메모리 소모를 피하려면 from_pipe를 사용하세요.
pipeline = AutoPipelineForImage2Image.from_pipe(pipeline_text2image).to("cuda")

init_image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/cat.png")
init_image = init_image.resize((512, 512))

prompt = "cat wizard, gandalf, lord of the rings, detailed, fantasy, cute, adorable, Pixar, Disney, 8k"

image = pipeline(prompt, image=init_image, strength=0.5, guidance_scale=0.0, num_inference_steps=2).images[0]
make_image_grid([init_image, image], rows=1, cols=2)
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/sdxl-turbo-img2img.png" alt="Image-to-image generation sample using SDXL Turbo"/>
</div>

## SDXL Turbo 속도 훨씬 더 빠르게 하기

- PyTorch 버전 2 이상을 사용하는 경우 UNet을 컴파일합니다. 첫 번째 추론 실행은 매우 느리지만 이후 실행은 훨씬 빨라집니다.

```py
pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
```

- 기본 VAE를 사용하는 경우, 각 생성 전후에 비용이 많이 드는 `dtype` 변환을 피하기 위해 `float32`로 유지하세요. 이 작업은 첫 생성 이전에 한 번만 수행하면 됩니다:

```py
pipe.upcast_vae()
```

또는, 커뮤니티 회원인 [`@madebyollin`](https://huggingface.co/madebyollin)이 만든 [16비트 VAE](https://huggingface.co/madebyollin/sdxl-vae-fp16-fix)를 사용할 수도 있으며, 이는 `float32`로 업캐스트할 필요가 없습니다.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/sdxl_turbo.md" />

### Stable Video Diffusion
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/svd.md

# Stable Video Diffusion


[Stable Video Diffusion (SVD)](https://huggingface.co/papers/2311.15127)은 입력 이미지에 맞춰 2~4초 분량의 고해상도(576x1024) 비디오를 생성할 수 있는 강력한 image-to-video 생성 모델입니다.

이 가이드에서는 SVD를 사용하여 이미지에서 짧은 동영상을 생성하는 방법을 설명합니다.

시작하기 전에 다음 라이브러리가 설치되어 있는지 확인하세요:

```py
!pip install -q -U diffusers transformers accelerate
```

이 모델에는 [SVD](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid)와 [SVD-XT](https://huggingface.co/stabilityai/stable-video-diffusion-img2vid-xt) 두 가지 종류가 있습니다. SVD 체크포인트는 14개의 프레임을 생성하도록 학습되었고, SVD-XT 체크포인트는 25개의 프레임을 생성하도록 파인튜닝되었습니다.

이 가이드에서는 SVD-XT 체크포인트를 사용합니다.

```python
import torch

from diffusers import StableVideoDiffusionPipeline
from diffusers.utils import load_image, export_to_video

pipe = StableVideoDiffusionPipeline.from_pretrained(
    "stabilityai/stable-video-diffusion-img2vid-xt", torch_dtype=torch.float16, variant="fp16"
)
pipe.enable_model_cpu_offload()

# Conditioning 이미지 불러오기
image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png")
image = image.resize((1024, 576))

generator = torch.manual_seed(42)
frames = pipe(image, decode_chunk_size=8, generator=generator).frames[0]

export_to_video(frames, "generated.mp4", fps=7)
```

<div class="flex gap-4">
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">"source image of a rocket"</figcaption>
  </div>
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/output_rocket.gif"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">"generated video from source image"</figcaption>
  </div>
</div>

## torch.compile

UNet을 [컴파일](../optimization/torch2.0#torchcompile)하면 메모리 사용량이 살짝 증가하지만, 20~25%의 속도 향상을 얻을 수 있습니다.

```diff
- pipe.enable_model_cpu_offload()
+ pipe.to("cuda")
+ pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
```

## 메모리 사용량 줄이기

비디오 생성은 기본적으로 배치 크기가 큰 text-to-image 생성과 유사하게 'num_frames'를 한 번에 생성하기 때문에 메모리 사용량이 매우 높습니다. 메모리 사용량을 줄이기 위해 추론 속도와 메모리 사용량을 절충하는 여러 가지 옵션이 있습니다:

- 모델 오프로링 활성화: 파이프라인의 각 구성 요소가 더 이상 필요하지 않을 때 CPU로 오프로드됩니다.
- Feed-forward chunking 활성화: feed-forward 레이어가 배치 크기가 큰 단일 feed-forward를 실행하는 대신 루프로 반복해서 실행됩니다.
- `decode_chunk_size` 감소: VAE가 프레임들을 한꺼번에 디코딩하는 대신 chunk 단위로 디코딩합니다. `decode_chunk_size=1`을 설정하면 한 번에 한 프레임씩 디코딩하고 최소한의 메모리만 사용하지만(GPU 메모리에 따라 이 값을 조정하는 것이 좋습니다), 동영상에 약간의 깜박임이 발생할 수 있습니다.

```diff
- pipe.enable_model_cpu_offload()
- frames = pipe(image, decode_chunk_size=8, generator=generator).frames[0]
+ pipe.enable_model_cpu_offload()
+ pipe.unet.enable_forward_chunking()
+ frames = pipe(image, decode_chunk_size=2, generator=generator, num_frames=25).frames[0]
```

이러한 모든 방법들을 사용하면 메모리 사용량이 8GAM VRAM보다 적을 것입니다.

## Micro-conditioning

Stable Diffusion Video는 또한 이미지 conditoning 외에도 micro-conditioning을 허용하므로 생성된 비디오를 더 잘 제어할 수 있습니다:

- `fps`: 생성된 비디오의 초당 프레임 수입니다.
- `motion_bucket_id`: 생성된 동영상에 사용할 모션 버킷 아이디입니다. 생성된 동영상의 모션을 제어하는 데 사용할 수 있습니다. 모션 버킷 아이디를 늘리면 생성되는 동영상의 모션이 증가합니다.
- `noise_aug_strength`: Conditioning 이미지에 추가되는 노이즈의 양입니다. 값이 클수록 비디오가 conditioning 이미지와 덜 유사해집니다. 이 값을 높이면 생성된 비디오의 움직임도 증가합니다.

예를 들어, 모션이 더 많은 동영상을 생성하려면 `motion_bucket_id` 및 `noise_aug_strength` micro-conditioning 파라미터를 사용합니다:

```python
import torch

from diffusers import StableVideoDiffusionPipeline
from diffusers.utils import load_image, export_to_video

pipe = StableVideoDiffusionPipeline.from_pretrained(
  "stabilityai/stable-video-diffusion-img2vid-xt", torch_dtype=torch.float16, variant="fp16"
)
pipe.enable_model_cpu_offload()

# Conditioning 이미지 불러오기
image = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/rocket.png")
image = image.resize((1024, 576))

generator = torch.manual_seed(42)
frames = pipe(image, decode_chunk_size=8, generator=generator, motion_bucket_id=180, noise_aug_strength=0.1).frames[0]
export_to_video(frames, "generated.mp4", fps=7)
```

![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/svd/output_rocket_with_conditions.gif)


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/svd.md" />

### 커스텀 파이프라인 불러오기
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/custom_pipeline_overview.md

# 커스텀 파이프라인 불러오기


커뮤니티 파이프라인은 논문에 명시된 원래의 구현체와 다른 형태로 구현된 모든 `DiffusionPipeline` 클래스를 의미합니다. (예를 들어, `StableDiffusionControlNetPipeline`는 ["Text-to-Image Generation with ControlNet Conditioning"](https://huggingface.co/papers/2302.05543) 해당) 이들은 추가 기능을 제공하거나 파이프라인의 원래 구현을 확장합니다.

[Speech to Image](https://github.com/huggingface/diffusers/tree/main/examples/community#speech-to-image) 또는 [Composable Stable Diffusion](https://github.com/huggingface/diffusers/tree/main/examples/community#composable-stable-diffusion) 과 같은 멋진 커뮤니티 파이프라인이 많이 있으며 [여기에서](https://github.com/huggingface/diffusers/tree/main/examples/community) 모든 공식 커뮤니티 파이프라인을 찾을 수 있습니다.

허브에서 커뮤니티 파이프라인을 로드하려면, 커뮤니티 파이프라인의 리포지토리 ID와 (파이프라인 가중치 및 구성 요소를 로드하려는) 모델의 리포지토리 ID를 인자로 전달해야 합니다. 예를 들어, 아래 예시에서는 `hf-internal-testing/diffusers-dummy-pipeline`에서 더미 파이프라인을 불러오고, `google/ddpm-cifar10-32`에서 파이프라인의 가중치와 컴포넌트들을 로드합니다.

> [!WARNING]
> 🔒 허깅 페이스 허브에서 커뮤니티 파이프라인을 불러오는 것은 곧 해당 코드가 안전하다고 신뢰하는 것입니다. 코드를 자동으로 불러오고 실행하기 앞서 반드시 온라인으로 해당 코드의 신뢰성을 검사하세요!

```py
from diffusers import DiffusionPipeline

pipeline = DiffusionPipeline.from_pretrained(
    "google/ddpm-cifar10-32", custom_pipeline="hf-internal-testing/diffusers-dummy-pipeline"
)
```

공식 커뮤니티 파이프라인을 불러오는 것은 비슷하지만, 공식 리포지토리 ID에서 가중치를 불러오는 것과 더불어 해당 파이프라인 내의 컴포넌트를 직접 지정하는 것 역시 가능합니다. 아래 예제를 보면 커뮤니티 [CLIP Guided Stable Diffusion](https://github.com/huggingface/diffusers/tree/main/examples/community#clip-guided-stable-diffusion) 파이프라인을 로드할 때, 해당 파이프라인에서 사용할 `clip_model` 컴포넌트와 `feature_extractor` 컴포넌트를 직접 설정하는 것을 확인할 수 있습니다.

```py
from diffusers import DiffusionPipeline
from transformers import CLIPImageProcessor, CLIPModel

clip_model_id = "laion/CLIP-ViT-B-32-laion2B-s34B-b79K"

feature_extractor = CLIPImageProcessor.from_pretrained(clip_model_id)
clip_model = CLIPModel.from_pretrained(clip_model_id)

pipeline = DiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    custom_pipeline="clip_guided_stable_diffusion",
    clip_model=clip_model,
    feature_extractor=feature_extractor,
)
```

커뮤니티 파이프라인에 대한 자세한 내용은 [커뮤니티 파이프라인](https://github.com/huggingface/diffusers/blob/main/docs/source/en/using-diffusers/custom_pipeline_examples) 가이드를 살펴보세요. 커뮤니티 파이프라인 등록에 관심이 있는 경우 [커뮤니티 파이프라인에 기여하는 방법](https://github.com/huggingface/diffusers/blob/main/docs/source/en/using-diffusers/contribute_pipeline)에 대한 가이드를 확인하세요 !

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/custom_pipeline_overview.md" />

### Unconditional 이미지 생성
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/unconditional_image_generation.md

# Unconditional 이미지 생성


Unconditional 이미지 생성은 비교적 간단한 작업입니다. 모델이 텍스트나 이미지와 같은 추가 조건 없이 이미 학습된 학습 데이터와 유사한 이미지만 생성합니다.

['DiffusionPipeline']은 추론을 위해 미리 학습된 diffusion 시스템을 사용하는 가장 쉬운 방법입니다.

먼저 ['DiffusionPipeline']의 인스턴스를 생성하고 다운로드할 파이프라인의 [체크포인트](https://huggingface.co/models?library=diffusers&sort=downloads)를 지정합니다. 허브의 🧨 diffusion 체크포인트 중 하나를 사용할 수 있습니다(사용할 체크포인트는 나비 이미지를 생성합니다).

> [!TIP]
> 💡 나만의 unconditional 이미지 생성 모델을 학습시키고 싶으신가요? 학습 가이드를 살펴보고 나만의 이미지를 생성하는 방법을 알아보세요.


이 가이드에서는 unconditional 이미지 생성에 ['DiffusionPipeline']과 [DDPM](https://huggingface.co/papers/2006.11239)을 사용합니다:

```python
 >>> from diffusers import DiffusionPipeline

 >>> generator = DiffusionPipeline.from_pretrained("anton-l/ddpm-butterflies-128")
```

[diffusion 파이프라인]은 모든 모델링, 토큰화, 스케줄링 구성 요소를 다운로드하고 캐시합니다. 이 모델은 약 14억 개의 파라미터로 구성되어 있기 때문에 GPU에서 실행할 것을 강력히 권장합니다. PyTorch에서와 마찬가지로 제너레이터 객체를 GPU로 옮길 수 있습니다:

```python
 >>> generator.to("cuda")
```

이제 제너레이터를 사용하여 이미지를 생성할 수 있습니다:

```python
 >>> image = generator().images[0]
```

출력은 기본적으로 [PIL.Image](https://pillow.readthedocs.io/en/stable/reference/Image.html?highlight=image#the-image-class) 객체로 감싸집니다.

다음을 호출하여 이미지를 저장할 수 있습니다:

```python
 >>> image.save("generated_image.png")
```

아래 스페이스(데모 링크)를 이용해 보고, 추론 단계의 매개변수를 자유롭게 조절하여 이미지 품질에 어떤 영향을 미치는지 확인해 보세요!

<iframe src="https://stevhliu-ddpm-butterflies-128.hf.space" frameborder="0" width="850" height="500"></iframe>


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/unconditional_image_generation.md" />

### 파이프라인, 모델, 스케줄러 불러오기
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/loading.md

# 파이프라인, 모델, 스케줄러 불러오기

기본적으로 diffusion 모델은 다양한 컴포넌트들(모델, 토크나이저, 스케줄러) 간의 복잡한 상호작용을 기반으로 동작합니다. 디퓨저스(Diffusers)는 이러한 diffusion 모델을 보다 쉽고 간편한 API로 제공하는 것을 목표로 설계되었습니다. `DiffusionPipeline`은 diffusion 모델이 갖는 복잡성을 하나의 파이프라인 API로 통합하고, 동시에 이를 구성하는 각각의 컴포넌트들을 태스크에 맞춰 유연하게 커스터마이징할 수 있도록 지원하고 있습니다.

diffusion 모델의 훈련과 추론에 필요한 모든 것은 `DiffusionPipeline.from_pretrained()` 메서드를 통해 접근할 수 있습니다. (이 말의 의미는 다음 단락에서 보다 자세하게 다뤄보도록 하겠습니다.)

이 문서에서는 설명할 내용은 다음과 같습니다.

* 허브를 통해 혹은 로컬로 파이프라인을 불러오는 법

* 파이프라인에 다른 컴포넌트들을 적용하는 법
* 오리지널 체크포인트가 아닌 variant를 불러오는 법  (variant란 기본으로 설정된 `fp32`가 아닌 다른  부동 소수점 타입(예: `fp16`)을 사용하거나 Non-EMA 가중치를 사용하는 체크포인트들을 의미합니다.)
* 모델과 스케줄러를 불러오는 법



## Diffusion 파이프라인

> [!TIP]
> 💡 `DiffusionPipeline` 클래스가 동작하는 방식에 보다 자세한 내용이 궁금하다면,  [DiffusionPipeline explained](#diffusionpipeline에-대해-알아보기) 섹션을 확인해보세요.

`DiffusionPipeline` 클래스는 diffusion 모델을 [허브](https://huggingface.co/models?library=diffusers)로부터 불러오는 가장 심플하면서 보편적인 방식입니다. `DiffusionPipeline.from_pretrained()` 메서드는 적합한 파이프라인 클래스를 자동으로 탐지하고, 필요한 구성요소(configuration)와 가중치(weight) 파일들을 다운로드하고 캐싱한 다음, 해당 파이프라인 인스턴스를 반환합니다.

```python
from diffusers import DiffusionPipeline

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipe = DiffusionPipeline.from_pretrained(repo_id)
```

물론 `DiffusionPipeline` 클래스를 사용하지 않고, 명시적으로 직접 해당 파이프라인 클래스를 불러오는 것도 가능합니다. 아래 예시 코드는 위 예시와 동일한 인스턴스를 반환합니다.

```python
from diffusers import StableDiffusionPipeline

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipe = StableDiffusionPipeline.from_pretrained(repo_id)
```

[CompVis/stable-diffusion-v1-4](https://huggingface.co/CompVis/stable-diffusion-v1-4)이나 [stable-diffusion-v1-5/stable-diffusion-v1-5](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) 같은 체크포인트들의 경우, 하나 이상의 다양한 태스크에 활용될 수 있습니다. (예를 들어 위의 두 체크포인트의 경우, text-to-image와 image-to-image에 모두 활용될 수 있습니다.)  만약 이러한 체크포인트들을 기본 설정 태스크가 아닌 다른 태스크에 활용하고자 한다면, 해당 태스크에 대응되는 파이프라인(task-specific pipeline)을 사용해야 합니다.

```python
from diffusers import StableDiffusionImg2ImgPipeline

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipe = StableDiffusionImg2ImgPipeline.from_pretrained(repo_id)
```



### 로컬 파이프라인

파이프라인을 로컬로 불러오고자 한다면, `git-lfs`를 사용하여 직접 체크포인트를 로컬 디스크에 다운로드 받아야 합니다. 아래의 명령어를 실행하면 `./stable-diffusion-v1-5`란 이름으로 폴더가 로컬디스크에 생성됩니다.

```bash
git lfs install
git clone https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5
```

그런 다음 해당 로컬 경로를 `from_pretrained()` 메서드에 전달합니다.

```python
from diffusers import DiffusionPipeline

repo_id = "./stable-diffusion-v1-5"
stable_diffusion = DiffusionPipeline.from_pretrained(repo_id)
```

위의 예시코드처럼 만약 `repo_id`가 로컬 패스(local path)라면, `from_pretrained()` 메서드는 이를 자동으로 감지하여 허브에서 파일을 다운로드하지 않습니다. 만약 로컬 디스크에 저장된 파이프라인 체크포인트가 최신 버전이 아닐 경우에도, 최신 버전을 다운로드하지 않고 기존 로컬 디스크에 저장된 체크포인트를 사용한다는 것을 의미합니다.



### 파이프라인 내부의 컴포넌트 교체하기

파이프라인 내부의 컴포넌트들은 호환 가능한 다른 컴포넌트로 교체될 수 있습니다. 이와 같은 컴포넌트 교체가 중요한 이유는 다음과 같습니다.

- 어떤 스케줄러를 사용할 것인가는 생성속도와 생성품질 간의 트레이드오프를 정의하는 중요한 요소입니다.
- diffusion 모델 내부의 컴포넌트들은 일반적으로 각각 독립적으로 훈련되기 때문에, 더 좋은 성능을 보여주는 컴포넌트가 있다면 그걸로 교체하는 식으로 성능을 향상시킬 수 있습니다.
- 파인 튜닝 단계에서는 일반적으로 UNet 혹은 텍스트 인코더와 같은 일부 컴포넌트들만 훈련하게 됩니다.

어떤 스케줄러들이 호환가능한지는 `compatibles` 속성을 통해 확인할 수 있습니다.

```python
from diffusers import DiffusionPipeline

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
stable_diffusion = DiffusionPipeline.from_pretrained(repo_id)
stable_diffusion.scheduler.compatibles
```

이번에는 `SchedulerMixin.from_pretrained()` 메서드를 사용해서, 기존 기본 스케줄러였던 `PNDMScheduler`를 보다 우수한 성능의 `EulerDiscreteScheduler`로 바꿔봅시다. 스케줄러를 로드할 때는 `subfolder` 인자를 통해, 해당 파이프라인의 리포지토리에서 [스케줄러에 관한 하위폴더](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5/tree/main/scheduler)를  명시해주어야 합니다.

그 다음 새롭게 생성한 `EulerDiscreteScheduler` 인스턴스를 `DiffusionPipeline`의 `scheduler` 인자에 전달합니다.

```python
from diffusers import DiffusionPipeline, EulerDiscreteScheduler, DPMSolverMultistepScheduler

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"

scheduler = EulerDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")

stable_diffusion = DiffusionPipeline.from_pretrained(repo_id, scheduler=scheduler)
```

### 세이프티 체커

스테이블 diffusion과 같은 diffusion 모델들은 유해한 이미지를 생성할 수도 있습니다. 이를 예방하기 위해 디퓨저스는 생성된 이미지의 유해성을 판단하는 [세이프티 체커(safety checker)](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py) 기능을 지원하고 있습니다. 만약 세이프티 체커의 사용을 원하지 않는다면, `safety_checker` 인자에 `None`을 전달해주시면 됩니다.

```python
from diffusers import DiffusionPipeline

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
stable_diffusion = DiffusionPipeline.from_pretrained(repo_id, safety_checker=None)
```

### 컴포넌트 재사용

복수의 파이프라인에 동일한 모델이 반복적으로 사용한다면, 굳이 해당 모델의 동일한 가중치를 중복으로 RAM에 불러올 필요는 없을 것입니다.  `components` 속성을 통해 파이프라인 내부의 컴포넌트들을 참조할 수 있는데, 이번 단락에서는 이를 통해 동일한 모델 가중치를 RAM에 중복으로 불러오는 것을 방지하는 법에 대해 알아보겠습니다.

```python
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
stable_diffusion_txt2img = StableDiffusionPipeline.from_pretrained(model_id)

components = stable_diffusion_txt2img.components
```

그 다음 위 예시 코드에서 선언한 `components` 변수를 다른 파이프라인에 전달함으로써, 모델의 가중치를 중복으로 RAM에 로딩하지 않고, 동일한 컴포넌트를 재사용할 수 있습니다.

```python
stable_diffusion_img2img = StableDiffusionImg2ImgPipeline(**components)
```

물론 각각의 컴포넌트들을 따로 따로 파이프라인에 전달할 수도 있습니다.  예를 들어 `stable_diffusion_txt2img` 파이프라인 안의 컴포넌트들 가운데서 세이프티 체커(`safety_checker`)와 피쳐 익스트랙터(`feature_extractor`)를 제외한 컴포넌트들만 `stable_diffusion_img2img` 파이프라인에서 재사용하는 방식 역시 가능합니다.

```python
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
stable_diffusion_txt2img = StableDiffusionPipeline.from_pretrained(model_id)
stable_diffusion_img2img = StableDiffusionImg2ImgPipeline(
    vae=stable_diffusion_txt2img.vae,
    text_encoder=stable_diffusion_txt2img.text_encoder,
    tokenizer=stable_diffusion_txt2img.tokenizer,
    unet=stable_diffusion_txt2img.unet,
    scheduler=stable_diffusion_txt2img.scheduler,
    safety_checker=None,
    feature_extractor=None,
    requires_safety_checker=False,
)
```

## Checkpoint variants

Variant란 일반적으로 다음과 같은 체크포인트들을 의미합니다.

-  `torch.float16`과 같이 정밀도는 더 낮지만, 용량 역시 더 작은 부동소수점 타입의 가중치를 사용하는 체크포인트. *(다만 이와 같은 variant의 경우, 추가적인 훈련과 CPU환경에서의 구동이 불가능합니다.)*
- Non-EMA 가중치를 사용하는 체크포인트. *(Non-EMA 가중치의 경우, 파인 튜닝 단계에서 사용하는 것이 권장되는데, 추론 단계에선 사용하지 않는 것이 권장됩니다.)*

> [!TIP]
> 💡 모델 구조는 동일하지만 서로 다른 학습 환경에서 서로 다른 데이터셋으로 학습된 체크포인트들이 있을 경우, 해당 체크포인트들은 variant 단계가 아닌 리포지토리 단계에서 분리되어 관리되어야 합니다. (즉, 해당 체크포인트들은 서로 다른 리포지토리에서 따로 관리되어야 합니다. 예시: `stable-diffusion-v1-4`, `stable-diffusion-v1-5`).

| **checkpoint type** | **weight name**                     | **argument for loading weights** |
| ------------------- | ----------------------------------- | -------------------------------- |
| original            | diffusion_pytorch_model.bin         |                                  |
| floating point      | diffusion_pytorch_model.fp16.bin    | `variant`, `torch_dtype`         |
| non-EMA             | diffusion_pytorch_model.non_ema.bin | `variant`                        |

variant를 로드할 때 2개의 중요한 argument가 있습니다.

* `torch_dtype`은 불러올 체크포인트의 부동소수점을 정의합니다. 예를 들어 `torch_dtype=torch.float16`을 명시함으로써 가중치의 부동소수점 타입을 `fl16`으로 변환할 수 있습니다. (만약 따로 설정하지 않을 경우, 기본값으로 `fp32` 타입의 가중치가 로딩됩니다.) 또한 `variant` 인자를 명시하지 않은 채로 체크포인트를 불러온 다음, 해당 체크포인트를 `torch_dtype=torch.float16` 인자를 통해 `fp16` 타입으로 변환하는 것 역시 가능합니다. 이 경우 기본으로 설정된 `fp32` 가중치가 먼저 다운로드되고, 해당 가중치들을 불러온 다음 `fp16` 타입으로 변환하게 됩니다.
* `variant` 인자는 리포지토리에서 어떤 variant를 불러올 것인가를 정의합니다. 가령  [`diffusers/stable-diffusion-variants`](https://huggingface.co/diffusers/stable-diffusion-variants/tree/main/unet) 리포지토리로부터 `non_ema` 체크포인트를 불러오고자 한다면, `variant="non_ema"` 인자를 전달해야 합니다.

```python
from diffusers import DiffusionPipeline

# load fp16 variant
stable_diffusion = DiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5", variant="fp16", torch_dtype=torch.float16
)
# load non_ema variant
stable_diffusion = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", variant="non_ema")
```

다른 부동소수점 타입의 가중치 혹은 non-EMA 가중치를 사용하는 체크포인트를 저장하기 위해서는, `DiffusionPipeline.save_pretrained()` 메서드를 사용해야 하며, 이 때 `variant` 인자를 명시해줘야 합니다. 원래의 체크포인트와 동일한 폴더에 variant를 저장해야 하며, 이렇게 하면 동일한 폴더에서 오리지널 체크포인트과 variant를 모두 불러올 수 있습니다.

```python
from diffusers import DiffusionPipeline

# save as fp16 variant
stable_diffusion.save_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", variant="fp16")
# save as non-ema variant
stable_diffusion.save_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", variant="non_ema")
```

만약 variant를 기존 폴더에 저장하지 않을 경우, `variant` 인자를 반드시 명시해야 합니다. 그렇게 하지 않을 경우 원래의 오리지널 체크포인트를 찾을 수 없게 되기 때문에 에러가 발생합니다.

```python
# 👎 this won't work
stable_diffusion = DiffusionPipeline.from_pretrained("./stable-diffusion-v1-5", torch_dtype=torch.float16)
# 👍 this works
stable_diffusion = DiffusionPipeline.from_pretrained(
    "./stable-diffusion-v1-5", variant="fp16", torch_dtype=torch.float16
)
```

### 모델 불러오기

모델들은 `ModelMixin.from_pretrained()` 메서드를 통해 불러올 수 있습니다. 해당 메서드는 최신 버전의 모델 가중치 파일과 설정 파일(configurations)을 다운로드하고 캐싱합니다. 만약 이러한 파일들이 최신 버전으로 로컬 캐시에 저장되어 있다면, `ModelMixin.from_pretrained()`는 굳이 해당 파일들을 다시 다운로드하지 않으며, 그저 캐시에 있는 최신 파일들을 재사용합니다.

모델은 `subfolder` 인자에 명시된 하위 폴더로부터 로드됩니다. 예를 들어 `stable-diffusion-v1-5/stable-diffusion-v1-5`의 UNet 모델의 가중치는 [`unet`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5/tree/main/unet) 폴더에 저장되어 있습니다.

```python
from diffusers import UNet2DConditionModel

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
model = UNet2DConditionModel.from_pretrained(repo_id, subfolder="unet")
```

혹은 [해당 모델의 리포지토리](https://huggingface.co/google/ddpm-cifar10-32/tree/main)로부터 다이렉트로 가져오는 것 역시 가능합니다.

```python
from diffusers import UNet2DModel

repo_id = "google/ddpm-cifar10-32"
model = UNet2DModel.from_pretrained(repo_id)
```

또한 앞서 봤던 `variant` 인자를 명시함으로써, Non-EMA나 `fp16`의 가중치를 가져오는 것 역시 가능합니다.

```python
from diffusers import UNet2DConditionModel

model = UNet2DConditionModel.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", subfolder="unet", variant="non-ema")
model.save_pretrained("./local-unet", variant="non-ema")
```

### 스케줄러

스케줄러들은 `SchedulerMixin.from_pretrained()` 메서드를 통해 불러올 수 있습니다. 모델과 달리 스케줄러는 별도의 가중치를 갖지 않으며, 따라서 당연히 별도의 학습과정을 요구하지 않습니다. 이러한 스케줄러들은 (해당 스케줄러 하위폴더의) configration 파일을 통해 정의됩니다.

여러개의 스케줄러를 불러온다고 해서 많은 메모리를 소모하는 것은 아니며, 다양한 스케줄러들에 동일한 스케줄러 configration을  적용하는 것 역시 가능합니다. 다음 예시 코드에서 불러오는 스케줄러들은 모두 `StableDiffusionPipeline`과 호환되는데, 이는 곧 해당 스케줄러들에 동일한 스케줄러 configration 파일을 적용할 수 있음을 의미합니다.

```python
from diffusers import StableDiffusionPipeline
from diffusers import (
    DDPMScheduler,
    DDIMScheduler,
    PNDMScheduler,
    LMSDiscreteScheduler,
    EulerDiscreteScheduler,
    EulerAncestralDiscreteScheduler,
    DPMSolverMultistepScheduler,
)

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"

ddpm = DDPMScheduler.from_pretrained(repo_id, subfolder="scheduler")
ddim = DDIMScheduler.from_pretrained(repo_id, subfolder="scheduler")
pndm = PNDMScheduler.from_pretrained(repo_id, subfolder="scheduler")
lms = LMSDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
euler_anc = EulerAncestralDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
euler = EulerDiscreteScheduler.from_pretrained(repo_id, subfolder="scheduler")
dpm = DPMSolverMultistepScheduler.from_pretrained(repo_id, subfolder="scheduler")

# replace `dpm` with any of `ddpm`, `ddim`, `pndm`, `lms`, `euler_anc`, `euler`
pipeline = StableDiffusionPipeline.from_pretrained(repo_id, scheduler=dpm)
```

### DiffusionPipeline에 대해 알아보기

클래스 메서드로서  `DiffusionPipeline.from_pretrained()`은 2가지를 담당합니다.

- 첫째로, `from_pretrained` 메서드는 최신 버전의 파이프라인을 다운로드하고, 캐시에 저장합니다. 이미 로컬 캐시에 최신 버전의 파이프라인이 저장되어 있다면, `DiffusionPipeline.from_pretrained()`은 해당 파일들을 다시 다운로드하지 않고, 로컬 캐시에 저장되어 있는 파이프라인을 불러옵니다.
-  `model_index.json` 파일을 통해 체크포인트에 대응되는 적합한 파이프라인 클래스로 불러옵니다.

파이프라인의 폴더 구조는 해당 파이프라인 클래스의 구조와 직접적으로 일치합니다. 예를 들어 `StableDiffusionPipeline` 클래스는 [`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) 리포지토리와 대응되는 구조를 갖습니다.

```python
from diffusers import DiffusionPipeline

repo_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipeline = DiffusionPipeline.from_pretrained(repo_id)
print(pipeline)
```

위의 코드 출력 결과를 확인해보면, `pipeline`은 `StableDiffusionPipeline`의 인스턴스이며, 다음과 같이 총 7개의 컴포넌트로 구성된다는 것을 알 수 있습니다.

- `"feature_extractor"`: `CLIPImageProcessor`의 인스턴스
- `"safety_checker"`: 유해한 컨텐츠를 스크리닝하기 위한 [컴포넌트](https://github.com/huggingface/diffusers/blob/e55687e1e15407f60f32242027b7bb8170e58266/src/diffusers/pipelines/stable_diffusion/safety_checker.py#L32)
- `"scheduler"`: `PNDMScheduler`의 인스턴스
- `"text_encoder"`: `CLIPTextModel`의 인스턴스
- `"tokenizer"`: a `CLIPTokenizer`의 인스턴스
- `"unet"`: `UNet2DConditionModel`의 인스턴스
- `"vae"` `AutoencoderKL`의 인스턴스

```json
StableDiffusionPipeline {
  "feature_extractor": [
    "transformers",
    "CLIPImageProcessor"
  ],
  "safety_checker": [
    "stable_diffusion",
    "StableDiffusionSafetyChecker"
  ],
  "scheduler": [
    "diffusers",
    "PNDMScheduler"
  ],
  "text_encoder": [
    "transformers",
    "CLIPTextModel"
  ],
  "tokenizer": [
    "transformers",
    "CLIPTokenizer"
  ],
  "unet": [
    "diffusers",
    "UNet2DConditionModel"
  ],
  "vae": [
    "diffusers",
    "AutoencoderKL"
  ]
}
```

파이프라인 인스턴스의 컴포넌트들을  [`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)의 폴더 구조와 비교해볼 경우, 각각의 컴포넌트마다 별도의 폴더가 있음을 확인할 수 있습니다.

```
.
├── feature_extractor
│   └── preprocessor_config.json
├── model_index.json
├── safety_checker
│   ├── config.json
│   └── pytorch_model.bin
├── scheduler
│   └── scheduler_config.json
├── text_encoder
│   ├── config.json
│   └── pytorch_model.bin
├── tokenizer
│   ├── merges.txt
│   ├── special_tokens_map.json
│   ├── tokenizer_config.json
│   └── vocab.json
├── unet
│   ├── config.json
│   ├── diffusion_pytorch_model.bin
└── vae
    ├── config.json
    ├── diffusion_pytorch_model.bin
```

또한 각각의 컴포넌트들을 파이프라인 인스턴스의 속성으로써 참조할 수 있습니다.

```py
pipeline.tokenizer
```

```python
CLIPTokenizer(
    name_or_path="/root/.cache/huggingface/hub/models--runwayml--stable-diffusion-v1-5/snapshots/39593d5650112b4cc580433f6b0435385882d819/tokenizer",
    vocab_size=49408,
    model_max_length=77,
    is_fast=False,
    padding_side="right",
    truncation_side="right",
    special_tokens={
        "bos_token": AddedToken("<|startoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
        "eos_token": AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
        "unk_token": AddedToken("<|endoftext|>", rstrip=False, lstrip=False, single_word=False, normalized=True),
        "pad_token": "<|endoftext|>",
    },
)
```

모든 파이프라인은 `model_index.json` 파일을 통해 `DiffusionPipeline`에 다음과 같은 정보를 전달합니다.

- `_class_name` 는 어떤 파이프라인 클래스를 사용해야 하는지에 대해 알려줍니다.
- `_diffusers_version`는 어떤 버전의 디퓨저스로 파이프라인 안의 모델들이 만들어졌는지를 알려줍니다.
- 그 다음은 각각의 컴포넌트들이 어떤 라이브러리의 어떤 클래스로 만들어졌는지에 대해 알려줍니다. (아래 예시에서 `"feature_extractor" : ["transformers", "CLIPImageProcessor"]`의 경우, `feature_extractor` 컴포넌트는 `transformers` 라이브러리의 `CLIPImageProcessor` 클래스를 통해 만들어졌다는 것을 의미합니다.)

```json
{
  "_class_name": "StableDiffusionPipeline",
  "_diffusers_version": "0.6.0",
  "feature_extractor": [
    "transformers",
    "CLIPImageProcessor"
  ],
  "safety_checker": [
    "stable_diffusion",
    "StableDiffusionSafetyChecker"
  ],
  "scheduler": [
    "diffusers",
    "PNDMScheduler"
  ],
  "text_encoder": [
    "transformers",
    "CLIPTextModel"
  ],
  "tokenizer": [
    "transformers",
    "CLIPTokenizer"
  ],
  "unet": [
    "diffusers",
    "UNet2DConditionModel"
  ],
  "vae": [
    "diffusers",
    "AutoencoderKL"
  ]
}
```



<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/loading.md" />

### Shap-E
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/shap-e.md

# Shap-E


Shap-E는 비디오 게임 개발, 인테리어 디자인, 건축에 사용할 수 있는 3D 에셋을 생성하기 위한 conditional 모델입니다. 대규모 3D 에셋 데이터셋을 학습되었고, 각 오브젝트의 더 많은 뷰를 렌더링하고 4K point cloud 대신 16K를 생성하도록 후처리합니다. Shap-E 모델은 두 단계로 학습됩니다:

1. 인코더가 3D 에셋의 포인트 클라우드와 렌더링된 뷰를 받아들이고 에셋을 나타내는 implicit functions의 파라미터를 출력합니다.
2. 인코더가 생성한 latents를 바탕으로 diffusion 모델을 훈련하여 neural radiance fields(NeRF) 또는 textured 3D 메시를 생성하여 다운스트림 애플리케이션에서 3D 에셋을 더 쉽게 렌더링하고 사용할 수 있도록 합니다.

이 가이드에서는 Shap-E를 사용하여 나만의 3D 에셋을 생성하는 방법을 보입니다!

시작하기 전에 다음 라이브러리가 설치되어 있는지 확인하세요:

```py
# Colab에서 필요한 라이브러리를 설치하기 위해 주석을 제외하세요
#!pip install -q diffusers transformers accelerate trimesh
```

## Text-to-3D

3D 객체의 gif를 생성하려면 텍스트 프롬프트를 `ShapEPipeline`에 전달합니다. 파이프라인은 3D 객체를 생성하는 데 사용되는 이미지 프레임 리스트를 생성합니다.

```py
import torch
from diffusers import ShapEPipeline

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

pipe = ShapEPipeline.from_pretrained("openai/shap-e", torch_dtype=torch.float16, variant="fp16")
pipe = pipe.to(device)

guidance_scale = 15.0
prompt = ["A firecracker", "A birthday cupcake"]

images = pipe(
    prompt,
    guidance_scale=guidance_scale,
    num_inference_steps=64,
    frame_size=256,
).images
```

이제 `export_to_gif()` 함수를 사용하여 이미지 프레임 리스트를 3D 객체의 gif로 변환합니다.

```py
from diffusers.utils import export_to_gif

export_to_gif(images[0], "firecracker_3d.gif")
export_to_gif(images[1], "cake_3d.gif")
```

<div class="flex gap-4">
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/shap_e/firecracker_out.gif"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">prompt = "A firecracker"</figcaption>
  </div>
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/shap_e/cake_out.gif"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">prompt = "A birthday cupcake"</figcaption>
  </div>
</div>

## Image-to-3D

다른 이미지로부터 3D 개체를 생성하려면 `ShapEImg2ImgPipeline`을 사용합니다. 기존 이미지를 사용하거나 완전히 새로운 이미지를 생성할 수 있습니다. [Kandinsky 2.1](../api/pipelines/kandinsky) 모델을 사용하여 새 이미지를 생성해 보겠습니다.

```py
from diffusers import DiffusionPipeline
import torch

prior_pipeline = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16, use_safetensors=True).to("cuda")
pipeline = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16, use_safetensors=True).to("cuda")

prompt = "A cheeseburger, white background"

image_embeds, negative_image_embeds = prior_pipeline(prompt, guidance_scale=1.0).to_tuple()
image = pipeline(
    prompt,
    image_embeds=image_embeds,
    negative_image_embeds=negative_image_embeds,
).images[0]

image.save("burger.png")
```

치즈버거를 `ShapEImg2ImgPipeline`에 전달하여 3D representation을 생성합니다.

```py
from PIL import Image
from diffusers import ShapEImg2ImgPipeline
from diffusers.utils import export_to_gif

pipe = ShapEImg2ImgPipeline.from_pretrained("openai/shap-e-img2img", torch_dtype=torch.float16, variant="fp16").to("cuda")

guidance_scale = 3.0
image = Image.open("burger.png").resize((256, 256))

images = pipe(
    image,
    guidance_scale=guidance_scale,
    num_inference_steps=64,
    frame_size=256,
).images

gif_path = export_to_gif(images[0], "burger_3d.gif")
```

<div class="flex gap-4">
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/shap_e/burger_in.png"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">cheeseburger</figcaption>
  </div>
  <div>
    <img class="rounded-xl" src="https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/shap_e/burger_out.gif"/>
    <figcaption class="mt-2 text-center text-sm text-gray-500">3D cheeseburger</figcaption>
  </div>
</div>

## 메시 생성하기

Shap-E는 다운스트림 애플리케이션에 렌더링할 textured 메시 출력을 생성할 수도 있는 유연한 모델입니다. 이 예제에서는 🤗 Datasets 라이브러리에서 [Dataset viewer](https://huggingface.co/docs/hub/datasets-viewer#dataset-preview)를 사용해 메시 시각화를 지원하는 `glb` 파일로 변환합니다.

`output_type` 매개변수를 `"mesh"`로 지정함으로써 `ShapEPipeline`과 `ShapEImg2ImgPipeline` 모두에 대한 메시 출력을 생성할 수 있습니다:

```py
import torch
from diffusers import ShapEPipeline

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

pipe = ShapEPipeline.from_pretrained("openai/shap-e", torch_dtype=torch.float16, variant="fp16")
pipe = pipe.to(device)

guidance_scale = 15.0
prompt = "A birthday cupcake"

images = pipe(prompt, guidance_scale=guidance_scale, num_inference_steps=64, frame_size=256, output_type="mesh").images
```

메시 출력을 `ply` 파일로 저장하려면 `export_to_ply()` 함수를 사용합니다:

> [!TIP]
> 선택적으로 `export_to_obj()` 함수를 사용하여 메시 출력을 `obj` 파일로 저장할 수 있습니다. 다양한 형식으로 메시 출력을 저장할 수 있어 다운스트림에서 더욱 유연하게 사용할 수 있습니다!

```py
from diffusers.utils import export_to_ply

ply_path = export_to_ply(images[0], "3d_cake.ply")
print(f"Saved to folder: {ply_path}")
```

그 다음 trimesh 라이브러리를 사용하여 `ply` 파일을 `glb` 파일로 변환할 수 있습니다:

```py
import trimesh

mesh = trimesh.load("3d_cake.ply")
mesh_export = mesh.export("3d_cake.glb", file_type="glb")
```

기본적으로 메시 출력은 아래쪽 시점에 초점이 맞춰져 있지만 회전 변환을 적용하여 기본 시점을 변경할 수 있습니다:

```py
import trimesh
import numpy as np

mesh = trimesh.load("3d_cake.ply")
rot = trimesh.transformations.rotation_matrix(-np.pi / 2, [1, 0, 0])
mesh = mesh.apply_transform(rot)
mesh_export = mesh.export("3d_cake.glb", file_type="glb")
```

메시 파일을 데이터셋 레포지토리에 업로드해 Dataset viewer로 시각화하세요!

<div class="flex justify-center">
    <img class="rounded-xl" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/3D-cake.gif"/>
</div>


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/shap-e.md" />

### Text-guided depth-to-image 생성
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/depth2img.md

# Text-guided depth-to-image 생성


`StableDiffusionDepth2ImgPipeline`을 사용하면 텍스트 프롬프트와 초기 이미지를 전달하여 새 이미지의 생성을 조절할 수 있습니다. 또한 이미지 구조를 보존하기 위해 `depth_map`을 전달할 수도 있습니다. `depth_map`이 제공되지 않으면 파이프라인은 통합된 [depth-estimation model](https://github.com/isl-org/MiDaS)을 통해 자동으로 깊이를 예측합니다.


먼저 `StableDiffusionDepth2ImgPipeline`의 인스턴스를 생성합니다:

```python
import torch
import requests
from PIL import Image

from diffusers import StableDiffusionDepth2ImgPipeline

pipe = StableDiffusionDepth2ImgPipeline.from_pretrained(
    "stabilityai/stable-diffusion-2-depth",
    torch_dtype=torch.float16,
).to("cuda")
```

이제 프롬프트를 파이프라인에 전달합니다. 특정 단어가 이미지 생성을 가이드 하는것을 방지하기 위해 `negative_prompt`를 전달할 수도 있습니다:

```python
url = "http://images.cocodataset.org/val2017/000000039769.jpg"
init_image = Image.open(requests.get(url, stream=True).raw)
prompt = "two tigers"
n_prompt = "bad, deformed, ugly, bad anatomy"
image = pipe(prompt=prompt, image=init_image, negative_prompt=n_prompt, strength=0.7).images[0]
image
```

| Input                                                                           | Output                                                                                                                                |
|---------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------|
| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/coco-cats.png" width="500"/> | <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/depth2img-tigers.png" width="500"/> |

아래의 Spaces를 가지고 놀며 depth map이 있는 이미지와 없는 이미지의 차이가 있는지 확인해 보세요!

<iframe
	src="https://radames-stable-diffusion-depth2img.hf.space"
	frameborder="0"
	width="850"
	height="500"
></iframe>


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/depth2img.md" />

### Textual inversion
https://huggingface.co/docs/diffusers/main/ko/using-diffusers/textual_inversion_inference.md

# Textual inversion


`StableDiffusionPipeline`은  textual-inversion을 지원하는데, 이는 몇 개의 샘플 이미지만으로 stable diffusion과 같은 모델이 새로운 컨셉을 학습할 수 있도록 하는 기법입니다. 이를 통해 생성된 이미지를 더 잘 제어하고 특정 컨셉에 맞게 모델을 조정할 수 있습니다. 커뮤니티에서 만들어진 컨셉들의 컬렉션은 [Stable Diffusion Conceptualizer](https://huggingface.co/spaces/sd-concepts-library/stable-diffusion-conceptualizer)를 통해 빠르게 사용해볼 수 있습니다.

이 가이드에서는 Stable Diffusion Conceptualizer에서 사전학습한 컨셉을 사용하여 textual-inversion으로 추론을 실행하는 방법을 보여드립니다. textual-inversion으로 모델에 새로운 컨셉을 학습시키는 데 관심이 있으시다면,  [Textual Inversion](./training/text_inversion)  훈련 가이드를 참조하세요.

Hugging Face 계정으로 로그인하세요:

```py
from huggingface_hub import notebook_login

notebook_login()
```

필요한 라이브러리를 불러오고 생성된 이미지를 시각화하기 위한 도우미 함수 `image_grid`를 만듭니다:

```py
import os
import torch

import PIL
from PIL import Image

from diffusers import StableDiffusionPipeline
from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer


def image_grid(imgs, rows, cols):
    assert len(imgs) == rows * cols

    w, h = imgs[0].size
    grid = Image.new("RGB", size=(cols * w, rows * h))
    grid_w, grid_h = grid.size

    for i, img in enumerate(imgs):
        grid.paste(img, box=(i % cols * w, i // cols * h))
    return grid
```

Stable Diffusion과 [Stable Diffusion Conceptualizer](https://huggingface.co/spaces/sd-concepts-library/stable-diffusion-conceptualizer)에서 사전학습된 컨셉을 선택합니다:

```py
pretrained_model_name_or_path = "stable-diffusion-v1-5/stable-diffusion-v1-5"
repo_id_embeds = "sd-concepts-library/cat-toy"
```

이제 파이프라인을 로드하고 사전학습된 컨셉을 파이프라인에 전달할 수 있습니다:

```py
pipeline = StableDiffusionPipeline.from_pretrained(pretrained_model_name_or_path, torch_dtype=torch.float16).to("cuda")

pipeline.load_textual_inversion(repo_id_embeds)
```

특별한 placeholder token '`<cat-toy>`'를 사용하여 사전학습된 컨셉으로 프롬프트를 만들고, 생성할 샘플의 수와 이미지 행의 수를 선택합니다:

```py
prompt = "a grafitti in a favela wall with a <cat-toy> on it"

num_samples = 2
num_rows = 2
```

그런 다음 파이프라인을 실행하고, 생성된 이미지들을 저장합니다. 그리고 처음에 만들었던 도우미 함수 `image_grid`를 사용하여 생성 결과들을 시각화합니다. 이 때 `num_inference_steps`와 `guidance_scale`과 같은 매개 변수들을 조정하여, 이것들이 이미지 품질에 어떠한 영향을 미치는지를 자유롭게 확인해보시기 바랍니다.

```py
all_images = []
for _ in range(num_rows):
    images = pipe(prompt, num_images_per_prompt=num_samples, num_inference_steps=50, guidance_scale=7.5).images
    all_images.extend(images)

grid = image_grid(all_images, num_samples, num_rows)
grid
```

<div class="flex justify-center">
    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/textual_inversion_inference.png">
</div>


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/using-diffusers/textual_inversion_inference.md" />

### Diffusion 모델 평가하기[[evaluating-diffusion-models]]
https://huggingface.co/docs/diffusers/main/ko/conceptual/evaluation.md

# Diffusion 모델 평가하기[[evaluating-diffusion-models]]

<a target="_blank" href="https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/evaluation.ipynb">
    <img src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/>
</a>

[Stable Diffusion](https://huggingface.co/docs/diffusers/stable_diffusion)와 같은 생성 모델의 평가는 주관적인 성격을 가지고 있습니다. 그러나 실무자와 연구자로서 우리는 종종 다양한 가능성 중에서 신중한 선택을 해야 합니다. 그래서 다양한 생성 모델 (GAN, Diffusion 등)을 사용할 때 어떻게 선택해야 할까요?

정성적인 평가는 모델의 이미지 품질에 대한 주관적인 평가이므로 오류가 발생할 수 있고 결정에 잘못된 영향을 미칠 수 있습니다. 반면, 정량적인 평가는 이미지 품질과 직접적인 상관관계를 갖지 않을 수 있습니다. 따라서 일반적으로 정성적 평가와 정량적 평가를 모두 고려하는 것이 더 강력한 신호를 제공하여 모델 선택에 도움이 됩니다.

이 문서에서는 Diffusion 모델을 평가하기 위한 정성적 및 정량적 방법에 대해 상세히 설명합니다. 정량적 방법에 대해서는 특히 `diffusers`와 함께 구현하는 방법에 초점을 맞추었습니다.

이 문서에서 보여진 방법들은 기반 생성 모델을 고정시키고 다양한 [노이즈 스케줄러](https://huggingface.co/docs/diffusers/main/en/api/schedulers/overview)를 평가하는 데에도 사용할 수 있습니다.

## 시나리오[[scenarios]]
다음과 같은 파이프라인을 사용하여 Diffusion 모델을 다룹니다:

- 텍스트로 안내된 이미지 생성 (예: [`StableDiffusionPipeline`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/text2img)).
- 입력 이미지에 추가로 조건을 건 텍스트로 안내된 이미지 생성 (예: [`StableDiffusionImg2ImgPipeline`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/img2img) 및 [`StableDiffusionInstructPix2PixPipeline`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/pix2pix)).
- 클래스 조건화된 이미지 생성 모델 (예: [`DiTPipeline`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/dit)).

## 정성적 평가[[qualitative-evaluation]]

정성적 평가는 일반적으로 생성된 이미지의 인간 평가를 포함합니다. 품질은 구성성, 이미지-텍스트 일치, 공간 관계 등과 같은 측면에서 측정됩니다. 일반적인 프롬프트는 주관적인 지표에 대한 일정한 기준을 제공합니다.
DrawBench와 PartiPrompts는 정성적인 벤치마킹에 사용되는 프롬프트 데이터셋입니다. DrawBench와 PartiPrompts는 각각 [Imagen](https://imagen.research.google/)과 [Parti](https://parti.research.google/)에서 소개되었습니다.

[Parti 공식 웹사이트](https://parti.research.google/)에서 다음과 같이 설명하고 있습니다:

> PartiPrompts (P2)는 이 작업의 일부로 공개되는 영어로 된 1600개 이상의 다양한 프롬프트 세트입니다. P2는 다양한 범주와 도전 측면에서 모델의 능력을 측정하는 데 사용할 수 있습니다.

![parti-prompts](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/parti-prompts.png)

PartiPrompts는 다음과 같은 열을 가지고 있습니다:

- 프롬프트 (Prompt)
- 프롬프트의 카테고리 (예: "Abstract", "World Knowledge" 등)
- 난이도를 반영한 챌린지 (예: "Basic", "Complex", "Writing & Symbols" 등)

이러한 벤치마크는 서로 다른 이미지 생성 모델을 인간 평가로 비교할 수 있도록 합니다.

이를 위해 🧨 Diffusers 팀은 **Open Parti Prompts**를 구축했습니다. 이는 Parti Prompts를 기반으로 한 커뮤니티 기반의 질적 벤치마크로, 최첨단 오픈 소스 확산 모델을 비교하는 데 사용됩니다:
- [Open Parti Prompts 게임](https://huggingface.co/spaces/OpenGenAI/open-parti-prompts): 10개의 parti prompt에 대해 4개의 생성된 이미지가 제시되며, 사용자는 프롬프트에 가장 적합한 이미지를 선택합니다.
- [Open Parti Prompts 리더보드](https://huggingface.co/spaces/OpenGenAI/parti-prompts-leaderboard): 현재 최고의 오픈 소스 diffusion 모델들을 서로 비교하는 리더보드입니다.

이미지를 수동으로 비교하려면, `diffusers`를 사용하여 몇가지 PartiPrompts를 어떻게 활용할 수 있는지 알아봅시다.

다음은 몇 가지 다른 도전에서 샘플링한 프롬프트를 보여줍니다: Basic, Complex, Linguistic Structures, Imagination, Writing & Symbols. 여기서는 PartiPrompts를 [데이터셋](https://huggingface.co/datasets/nateraw/parti-prompts)으로 사용합니다.

```python
from datasets import load_dataset

# prompts = load_dataset("nateraw/parti-prompts", split="train")
# prompts = prompts.shuffle()
# sample_prompts = [prompts[i]["Prompt"] for i in range(5)]

# Fixing these sample prompts in the interest of reproducibility.
sample_prompts = [
    "a corgi",
    "a hot air balloon with a yin-yang symbol, with the moon visible in the daytime sky",
    "a car with no windows",
    "a cube made of porcupine",
    'The saying "BE EXCELLENT TO EACH OTHER" written on a red brick wall with a graffiti image of a green alien wearing a tuxedo. A yellow fire hydrant is on a sidewalk in the foreground.',
]
```
이제 이런 프롬프트를 사용하여 Stable Diffusion ([v1-4 checkpoint](https://huggingface.co/CompVis/stable-diffusion-v1-4))를 사용한 이미지 생성을 할 수 있습니다 :

```python
import torch

seed = 0
generator = torch.manual_seed(seed)

images = sd_pipeline(sample_prompts, num_images_per_prompt=1, generator=generator).images
```

![parti-prompts-14](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/parti-prompts-14.png)


`num_images_per_prompt`를 설정하여 동일한 프롬프트에 대해 다른 이미지를 비교할 수도 있습니다. 다른 체크포인트([v1-5](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5))로 동일한 파이프라인을 실행하면 다음과 같은 결과가 나옵니다:

![parti-prompts-15](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/parti-prompts-15.png)


다양한 모델을 사용하여 모든 프롬프트에서 생성된 여러 이미지들이 생성되면 (평가 과정에서) 이러한 결과물들은 사람 평가자들에게 점수를 매기기 위해 제시됩니다. DrawBench와 PartiPrompts 벤치마크에 대한 자세한 내용은 각각의 논문을 참조하십시오.

> [!TIP]
> 모델이 훈련 중일 때 추론 샘플을 살펴보는 것은 훈련 진행 상황을 측정하는 데 유용합니다. [훈련 스크립트](https://github.com/huggingface/diffusers/tree/main/examples/)에서는 TensorBoard와 Weights & Biases에 대한 추가 지원과 함께 이 유틸리티를 지원합니다.

## 정량적 평가[[quantitative-evaluation]]

이 섹션에서는 세 가지 다른 확산 파이프라인을 평가하는 방법을 안내합니다:

- CLIP 점수
- CLIP 방향성 유사도
- FID

### 텍스트 안내 이미지 생성[[text-guided-image-generation]]

[CLIP 점수](https://huggingface.co/papers/2104.08718)는 이미지-캡션 쌍의 호환성을 측정합니다. 높은 CLIP 점수는 높은 호환성🔼을 나타냅니다. CLIP 점수는 이미지와 캡션 사이의 의미적 유사성으로 생각할 수도 있습니다. CLIP 점수는 인간 판단과 높은 상관관계를 가지고 있습니다.

`StableDiffusionPipeline`을 일단 로드해봅시다:

```python
from diffusers import StableDiffusionPipeline
import torch

model_ckpt = "CompVis/stable-diffusion-v1-4"
sd_pipeline = StableDiffusionPipeline.from_pretrained(model_ckpt, torch_dtype=torch.float16).to("cuda")
```

여러 개의 프롬프트를 사용하여 이미지를 생성합니다:

```python
prompts = [
    "a photo of an astronaut riding a horse on mars",
    "A high tech solarpunk utopia in the Amazon rainforest",
    "A pikachu fine dining with a view to the Eiffel Tower",
    "A mecha robot in a favela in expressionist style",
    "an insect robot preparing a delicious meal",
    "A small cabin on top of a snowy mountain in the style of Disney, artstation",
]

images = sd_pipeline(prompts, num_images_per_prompt=1, output_type="np").images

print(images.shape)
# (6, 512, 512, 3)
```

그러고 나서 CLIP 점수를 계산합니다.

```python
from torchmetrics.functional.multimodal import clip_score
from functools import partial

clip_score_fn = partial(clip_score, model_name_or_path="openai/clip-vit-base-patch16")

def calculate_clip_score(images, prompts):
    images_int = (images * 255).astype("uint8")
    clip_score = clip_score_fn(torch.from_numpy(images_int).permute(0, 3, 1, 2), prompts).detach()
    return round(float(clip_score), 4)

sd_clip_score = calculate_clip_score(images, prompts)
print(f"CLIP score: {sd_clip_score}")
# CLIP score: 35.7038
```

위의 예제에서는 각 프롬프트 당 하나의 이미지를 생성했습니다. 만약 프롬프트 당 여러 이미지를 생성한다면, 프롬프트 당 생성된 이미지의 평균 점수를 사용해야 합니다.

이제 `StableDiffusionPipeline`과 호환되는 두 개의 체크포인트를 비교하려면, 파이프라인을 호출할 때 generator를 전달해야 합니다. 먼저, 고정된 시드로 [v1-4 Stable Diffusion 체크포인트](https://huggingface.co/CompVis/stable-diffusion-v1-4)를 사용하여 이미지를 생성합니다:

```python
seed = 0
generator = torch.manual_seed(seed)

images = sd_pipeline(prompts, num_images_per_prompt=1, generator=generator, output_type="np").images
```

그런 다음 [v1-5 checkpoint](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)를 로드하여 이미지를 생성합니다:

```python
model_ckpt_1_5 = "stable-diffusion-v1-5/stable-diffusion-v1-5"
sd_pipeline_1_5 = StableDiffusionPipeline.from_pretrained(model_ckpt_1_5, torch_dtype=weight_dtype).to(device)

images_1_5 = sd_pipeline_1_5(prompts, num_images_per_prompt=1, generator=generator, output_type="np").images
```

그리고 마지막으로 CLIP 점수를 비교합니다:

```python
sd_clip_score_1_4 = calculate_clip_score(images, prompts)
print(f"CLIP Score with v-1-4: {sd_clip_score_1_4}")
# CLIP Score with v-1-4: 34.9102

sd_clip_score_1_5 = calculate_clip_score(images_1_5, prompts)
print(f"CLIP Score with v-1-5: {sd_clip_score_1_5}")
# CLIP Score with v-1-5: 36.2137
```

[v1-5](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) 체크포인트가 이전 버전보다 더 나은 성능을 보이는 것 같습니다. 그러나 CLIP 점수를 계산하기 위해 사용한 프롬프트의 수가 상당히 적습니다. 보다 실용적인 평가를 위해서는 이 수를 훨씬 높게 설정하고, 프롬프트를 다양하게 사용해야 합니다.

> [!WARNING]
> 이 점수에는 몇 가지 제한 사항이 있습니다. 훈련 데이터셋의 캡션은 웹에서 크롤링되어 이미지와 관련된 `alt` 및 유사한 태그에서 추출되었습니다. 이들은 인간이 이미지를 설명하는 데 사용할 수 있는 것과 일치하지 않을 수 있습니다. 따라서 여기서는 몇 가지 프롬프트를 "엔지니어링"해야 했습니다.

### 이미지 조건화된 텍스트-이미지 생성[[image-conditioned-text-to-image-generation]]

이 경우, 생성 파이프라인을 입력 이미지와 텍스트 프롬프트로 조건화합니다. `StableDiffusionInstructPix2PixPipeline`을 예로 들어보겠습니다. 이는 편집 지시문을 입력 프롬프트로 사용하고 편집할 입력 이미지를 사용합니다.

다음은 하나의 예시입니다:

![edit-instruction](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/edit-instruction.png)

모델을 평가하는 한 가지 전략은 두 이미지 캡션 간의 변경과([CLIP-Guided Domain Adaptation of Image Generators](https://huggingface.co/papers/2108.00946)에서 보여줍니다) 함께 두 이미지 사이의 변경의 일관성을 측정하는 것입니다 ([CLIP](https://huggingface.co/docs/transformers/model_doc/clip) 공간에서). 이를 "**CLIP 방향성 유사성**"이라고 합니다.

- 캡션 1은 편집할 이미지 (이미지 1)에 해당합니다.
- 캡션 2는 편집된 이미지 (이미지 2)에 해당합니다. 편집 지시를 반영해야 합니다.

다음은 그림으로 된 개요입니다:

![edit-consistency](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/edit-consistency.png)

우리는 이 측정 항목을 구현하기 위해 미니 데이터 세트를 준비했습니다. 먼저 데이터 세트를 로드해 보겠습니다.

```python
from datasets import load_dataset

dataset = load_dataset("sayakpaul/instructpix2pix-demo", split="train")
dataset.features
```

```bash
{'input': Value(dtype='string', id=None),
 'edit': Value(dtype='string', id=None),
 'output': Value(dtype='string', id=None),
 'image': Image(decode=True, id=None)}
```

여기에는 다음과 같은 항목이 있습니다:

- `input`은 `image`에 해당하는 캡션입니다.
- `edit`은 편집 지시사항을 나타냅니다.
- `output`은 `edit` 지시사항을 반영한 수정된 캡션입니다.

샘플을 살펴보겠습니다.

```python
idx = 0
print(f"Original caption: {dataset[idx]['input']}")
print(f"Edit instruction: {dataset[idx]['edit']}")
print(f"Modified caption: {dataset[idx]['output']}")
```

```bash
Original caption: 2. FAROE ISLANDS: An archipelago of 18 mountainous isles in the North Atlantic Ocean between Norway and Iceland, the Faroe Islands has 'everything you could hope for', according to Big 7 Travel. It boasts 'crystal clear waterfalls, rocky cliffs that seem to jut out of nowhere and velvety green hills'
Edit instruction: make the isles all white marble
Modified caption: 2. WHITE MARBLE ISLANDS: An archipelago of 18 mountainous white marble isles in the North Atlantic Ocean between Norway and Iceland, the White Marble Islands has 'everything you could hope for', according to Big 7 Travel. It boasts 'crystal clear waterfalls, rocky cliffs that seem to jut out of nowhere and velvety green hills'
```

다음은 이미지입니다:

```python
dataset[idx]["image"]
```

![edit-dataset](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/edit-dataset.png)

먼저 편집 지시사항을 사용하여 데이터 세트의 이미지를 편집하고 방향 유사도를 계산합니다.

`StableDiffusionInstructPix2PixPipeline`를 먼저 로드합니다:

```python
from diffusers import StableDiffusionInstructPix2PixPipeline

instruct_pix2pix_pipeline = StableDiffusionInstructPix2PixPipeline.from_pretrained(
    "timbrooks/instruct-pix2pix", torch_dtype=torch.float16
).to(device)
```

이제 편집을 수행합니다:

```python
import numpy as np


def edit_image(input_image, instruction):
    image = instruct_pix2pix_pipeline(
        instruction,
        image=input_image,
        output_type="np",
        generator=generator,
    ).images[0]
    return image

input_images = []
original_captions = []
modified_captions = []
edited_images = []

for idx in range(len(dataset)):
    input_image = dataset[idx]["image"]
    edit_instruction = dataset[idx]["edit"]
    edited_image = edit_image(input_image, edit_instruction)

    input_images.append(np.array(input_image))
    original_captions.append(dataset[idx]["input"])
    modified_captions.append(dataset[idx]["output"])
    edited_images.append(edited_image)
```
방향 유사도를 계산하기 위해서는 먼저 CLIP의 이미지와 텍스트 인코더를 로드합니다:

```python
from transformers import (
    CLIPTokenizer,
    CLIPTextModelWithProjection,
    CLIPVisionModelWithProjection,
    CLIPImageProcessor,
)

clip_id = "openai/clip-vit-large-patch14"
tokenizer = CLIPTokenizer.from_pretrained(clip_id)
text_encoder = CLIPTextModelWithProjection.from_pretrained(clip_id).to(device)
image_processor = CLIPImageProcessor.from_pretrained(clip_id)
image_encoder = CLIPVisionModelWithProjection.from_pretrained(clip_id).to(device)
```

주목할 점은 특정한 CLIP 체크포인트인 `openai/clip-vit-large-patch14`를 사용하고 있다는 것입니다. 이는 Stable Diffusion 사전 훈련이 이 CLIP 변형체와 함께 수행되었기 때문입니다. 자세한 내용은 [문서](https://huggingface.co/docs/transformers/model_doc/clip)를 참조하세요.

다음으로, 방향성 유사도를 계산하기 위해 PyTorch의 `nn.Module`을 준비합니다:

```python
import torch.nn as nn
import torch.nn.functional as F


class DirectionalSimilarity(nn.Module):
    def __init__(self, tokenizer, text_encoder, image_processor, image_encoder):
        super().__init__()
        self.tokenizer = tokenizer
        self.text_encoder = text_encoder
        self.image_processor = image_processor
        self.image_encoder = image_encoder

    def preprocess_image(self, image):
        image = self.image_processor(image, return_tensors="pt")["pixel_values"]
        return {"pixel_values": image.to(device)}

    def tokenize_text(self, text):
        inputs = self.tokenizer(
            text,
            max_length=self.tokenizer.model_max_length,
            padding="max_length",
            truncation=True,
            return_tensors="pt",
        )
        return {"input_ids": inputs.input_ids.to(device)}

    def encode_image(self, image):
        preprocessed_image = self.preprocess_image(image)
        image_features = self.image_encoder(**preprocessed_image).image_embeds
        image_features = image_features / image_features.norm(dim=1, keepdim=True)
        return image_features

    def encode_text(self, text):
        tokenized_text = self.tokenize_text(text)
        text_features = self.text_encoder(**tokenized_text).text_embeds
        text_features = text_features / text_features.norm(dim=1, keepdim=True)
        return text_features

    def compute_directional_similarity(self, img_feat_one, img_feat_two, text_feat_one, text_feat_two):
        sim_direction = F.cosine_similarity(img_feat_two - img_feat_one, text_feat_two - text_feat_one)
        return sim_direction

    def forward(self, image_one, image_two, caption_one, caption_two):
        img_feat_one = self.encode_image(image_one)
        img_feat_two = self.encode_image(image_two)
        text_feat_one = self.encode_text(caption_one)
        text_feat_two = self.encode_text(caption_two)
        directional_similarity = self.compute_directional_similarity(
            img_feat_one, img_feat_two, text_feat_one, text_feat_two
        )
        return directional_similarity
```

이제 `DirectionalSimilarity`를 사용해 보겠습니다.

```python
dir_similarity = DirectionalSimilarity(tokenizer, text_encoder, image_processor, image_encoder)
scores = []

for i in range(len(input_images)):
    original_image = input_images[i]
    original_caption = original_captions[i]
    edited_image = edited_images[i]
    modified_caption = modified_captions[i]

    similarity_score = dir_similarity(original_image, edited_image, original_caption, modified_caption)
    scores.append(float(similarity_score.detach().cpu()))

print(f"CLIP directional similarity: {np.mean(scores)}")
# CLIP directional similarity: 0.0797976553440094
```

CLIP 점수와 마찬가지로, CLIP 방향 유사성이 높을수록 좋습니다.

`StableDiffusionInstructPix2PixPipeline`은 `image_guidance_scale`과 `guidance_scale`이라는 두 가지 인자를 노출시킵니다. 이 두 인자를 조정하여 최종 편집된 이미지의 품질을 제어할 수 있습니다. 이 두 인자의 영향을 실험해보고 방향 유사성에 미치는 영향을 확인해보기를 권장합니다.

이러한 메트릭의 개념을 확장하여 원본 이미지와 편집된 버전의 유사성을 측정할 수 있습니다. 이를 위해 `F.cosine_similarity(img_feat_two, img_feat_one)`을 사용할 수 있습니다. 이러한 종류의 편집에서는 이미지의 주요 의미가 최대한 보존되어야 합니다. 즉, 높은 유사성 점수를 얻어야 합니다.

[`StableDiffusionPix2PixZeroPipeline`](https://huggingface.co/docs/diffusers/main/en/api/pipelines/pix2pix_zero#diffusers.StableDiffusionPix2PixZeroPipeline)와 같은 유사한 파이프라인에도 이러한 메트릭을 사용할 수 있습니다.

> [!TIP]
> CLIP 점수와 CLIP 방향 유사성 모두 CLIP 모델에 의존하기 때문에 평가가 편향될 수 있습니다

***IS, FID (나중에 설명할 예정), 또는 KID와 같은 메트릭을 확장하는 것은 어려울 수 있습니다***. 평가 중인 모델이 대규모 이미지 캡셔닝 데이터셋 (예: [LAION-5B 데이터셋](https://laion.ai/blog/laion-5b/))에서 사전 훈련되었을 때 이는 문제가 될 수 있습니다. 왜냐하면 이러한 메트릭의 기반에는 중간 이미지 특징을 추출하기 위해 ImageNet-1k 데이터셋에서 사전 훈련된 InceptionNet이 사용되기 때문입니다. Stable Diffusion의 사전 훈련 데이터셋은 InceptionNet의 사전 훈련 데이터셋과 겹치는 부분이 제한적일 수 있으므로 따라서 여기에는 좋은 후보가 아닙니다.

***위의 메트릭을 사용하면 클래스 조건이 있는 모델을 평가할 수 있습니다. 예를 들어, [DiT](https://huggingface.co/docs/diffusers/main/en/api/pipelines/dit). 이는 ImageNet-1k 클래스에 조건을 걸고 사전 훈련되었습니다.***

### 클래스 조건화 이미지 생성[[class-conditioned-image-generation]]

클래스 조건화 생성 모델은 일반적으로 [ImageNet-1k](https://huggingface.co/datasets/imagenet-1k)와 같은 클래스 레이블이 지정된 데이터셋에서 사전 훈련됩니다. 이러한 모델을 평가하는 인기있는 지표에는 Fréchet Inception Distance (FID), Kernel Inception Distance (KID) 및 Inception Score (IS)가 있습니다. 이 문서에서는 FID ([Heusel et al.](https://huggingface.co/papers/1706.08500))에 초점을 맞추고 있습니다. [`DiTPipeline`](https://huggingface.co/docs/diffusers/api/pipelines/dit)을 사용하여 FID를 계산하는 방법을 보여줍니다. 이는 내부적으로 [DiT 모델](https://huggingface.co/papers/2212.09748)을 사용합니다.

FID는 두 개의 이미지 데이터셋이 얼마나 유사한지를 측정하는 것을 목표로 합니다. [이 자료](https://mmgeneration.readthedocs.io/en/latest/quick_run.html#fid)에 따르면:

> Fréchet Inception Distance는 두 개의 이미지 데이터셋 간의 유사성을 측정하는 지표입니다. 시각적 품질에 대한 인간 판단과 잘 상관되는 것으로 나타났으며, 주로 생성적 적대 신경망의 샘플 품질을 평가하는 데 사용됩니다. FID는 Inception 네트워크의 특징 표현에 맞게 적합한 두 개의 가우시안 사이의 Fréchet 거리를 계산하여 구합니다.

이 두 개의 데이터셋은 실제 이미지 데이터셋과 가짜 이미지 데이터셋(우리의 경우 생성된 이미지)입니다. FID는 일반적으로 두 개의 큰 데이터셋으로 계산됩니다. 그러나 이 문서에서는 두 개의 미니 데이터셋으로 작업할 것입니다.

먼저 ImageNet-1k 훈련 세트에서 몇 개의 이미지를 다운로드해 봅시다:

```python
from zipfile import ZipFile
import requests


def download(url, local_filepath):
    r = requests.get(url)
    with open(local_filepath, "wb") as f:
        f.write(r.content)
    return local_filepath

dummy_dataset_url = "https://hf.co/datasets/sayakpaul/sample-datasets/resolve/main/sample-imagenet-images.zip"
local_filepath = download(dummy_dataset_url, dummy_dataset_url.split("/")[-1])

with ZipFile(local_filepath, "r") as zipper:
    zipper.extractall(".")
```

```python
from PIL import Image
import os

dataset_path = "sample-imagenet-images"
image_paths = sorted([os.path.join(dataset_path, x) for x in os.listdir(dataset_path)])

real_images = [np.array(Image.open(path).convert("RGB")) for path in image_paths]
```

다음은 ImageNet-1k classes의 이미지 10개입니다 : "cassette_player", "chain_saw" (x2), "church", "gas_pump" (x3), "parachute" (x2), 그리고 "tench".

<p align="center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/real-images.png" alt="real-images"><br>
    <em>Real images.</em>
</p>

이제 이미지가 로드되었으므로 이미지에 가벼운 전처리를 적용하여 FID 계산에 사용해 보겠습니다.

```python
from torchvision.transforms import functional as F


def preprocess_image(image):
    image = torch.tensor(image).unsqueeze(0)
    image = image.permute(0, 3, 1, 2) / 255.0
    return F.center_crop(image, (256, 256))

real_images = torch.cat([preprocess_image(image) for image in real_images])
print(real_images.shape)
# torch.Size([10, 3, 256, 256])
```

이제 위에서 언급한 클래스에 따라 조건화 된 이미지를 생성하기 위해 [`DiTPipeline`](https://huggingface.co/docs/diffusers/api/pipelines/dit)를 로드합니다.

```python
from diffusers import DiTPipeline, DPMSolverMultistepScheduler

dit_pipeline = DiTPipeline.from_pretrained("facebook/DiT-XL-2-256", torch_dtype=torch.float16)
dit_pipeline.scheduler = DPMSolverMultistepScheduler.from_config(dit_pipeline.scheduler.config)
dit_pipeline = dit_pipeline.to("cuda")

words = [
    "cassette player",
    "chainsaw",
    "chainsaw",
    "church",
    "gas pump",
    "gas pump",
    "gas pump",
    "parachute",
    "parachute",
    "tench",
]

class_ids = dit_pipeline.get_label_ids(words)
output = dit_pipeline(class_labels=class_ids, generator=generator, output_type="np")

fake_images = output.images
fake_images = torch.tensor(fake_images)
fake_images = fake_images.permute(0, 3, 1, 2)
print(fake_images.shape)
# torch.Size([10, 3, 256, 256])
```

이제 [`torchmetrics`](https://torchmetrics.readthedocs.io/)를 사용하여 FID를 계산할 수 있습니다.

```python
from torchmetrics.image.fid import FrechetInceptionDistance

fid = FrechetInceptionDistance(normalize=True)
fid.update(real_images, real=True)
fid.update(fake_images, real=False)

print(f"FID: {float(fid.compute())}")
# FID: 177.7147216796875
```

FID는 낮을수록 좋습니다. 여러 가지 요소가 FID에 영향을 줄 수 있습니다:

- 이미지의 수 (실제 이미지와 가짜 이미지 모두)
- diffusion 과정에서 발생하는 무작위성
- diffusion 과정에서의 추론 단계 수
- diffusion 과정에서 사용되는 스케줄러

마지막 두 가지 요소에 대해서는, 다른 시드와 추론 단계에서 평가를 실행하고 평균 결과를 보고하는 것은 좋은 실천 방법입니다

> [!WARNING]
> FID 결과는 많은 요소에 의존하기 때문에 취약할 수 있습니다:
>
> * 계산 중 사용되는 특정 Inception 모델.
> * 계산의 구현 정확도.
> * 이미지 형식 (PNG 또는 JPG에서 시작하는 경우가 다릅니다).
>
> 이러한 사항을 염두에 두면, FID는 유사한 실행을 비교할 때 가장 유용하지만, 저자가 FID 측정 코드를 주의 깊게 공개하지 않는 한 논문 결과를 재현하기는 어렵습니다.
>
> 이러한 사항은 KID 및 IS와 같은 다른 관련 메트릭에도 적용됩니다.

마지막 단계로, `fake_images`를 시각적으로 검사해 봅시다.

<p align="center">
    <img src="https://huggingface.co/datasets/diffusers/docs-images/resolve/main/evaluation_diffusion_models/fake-images.png" alt="fake-images"><br>
    <em>Fake images.</em>
</p>

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/conceptual/evaluation.md" />

### 철학 [[philosophy]]
https://huggingface.co/docs/diffusers/main/ko/conceptual/philosophy.md

# 철학 [[philosophy]]

🧨 Diffusers는 다양한 모달리티에서 **최신의** 사전 훈련된 diffusion 모델을 제공합니다.
그 목적은 추론과 훈련을 위한 **모듈식 툴박스**로 사용되는 것입니다.

저희는 시간이 지나도 변치 않는 라이브러리를 구축하는 것을 목표로 하기에 API 설계를 매우 중요하게 생각합니다.

간단히 말해서, Diffusers는 PyTorch를 자연스럽게 확장할 수 있도록 만들어졌습니다. 따라서 대부분의 설계 선택은 [PyTorch의 설계 원칙](https://pytorch.org/docs/stable/community/design.html#pytorch-design-philosophy)에 기반합니다. 이제 가장 중요한 것들을 살펴보겠습니다:

## 성능보다는 사용성을 [[usability-over-performance]]

- Diffusers는 다양한 성능 향상 기능이 내장되어 있지만 (자세한 내용은 [메모리와 속도](https://huggingface.co/docs/diffusers/optimization/fp16) 참조), 모델은 항상 가장 높은 정밀도와 최소한의 최적화로 로드됩니다. 따라서 사용자가 별도로 정의하지 않는 한 기본적으로 diffusion 파이프라인은 항상 float32 정밀도로 CPU에 인스턴스화됩니다. 이는 다양한 플랫폼과 가속기에서의 사용성을 보장하며, 라이브러리를 실행하기 위해 복잡한 설치가 필요하지 않다는 것을 의미합니다.
- Diffusers는 **가벼운** 패키지를 지향하기 때문에 필수 종속성은 거의 없지만 성능을 향상시킬 수 있는 많은 선택적 종속성이 있습니다 (`accelerate`, `safetensors`, `onnx` 등). 저희는 라이브러리를 가능한 한 가볍게 유지하여 다른 패키지에 대한 종속성 걱정이 없도록 노력하고 있습니다.
- Diffusers는 간결하고 이해하기 쉬운 코드를 선호합니다. 이는 람다 함수나 고급 PyTorch 연산자와 같은 압축된 코드 구문을 자주 사용하지 않는 것을 의미합니다.

## 쉬움보다는 간단함을 [[simple-over-easy]]

PyTorch에서는 **명시적인 것이 암시적인 것보다 낫다**와 **단순한 것이 복잡한 것보다 낫다**라고 말합니다. 이 설계 철학은 라이브러리의 여러 부분에 반영되어 있습니다:
- [`DiffusionPipeline.to`](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.to)와 같은 메소드를 사용하여 사용자가 장치 관리를 할 수 있도록 PyTorch의 API를 따릅니다.
- 잘못된 입력을 조용히 수정하는 대신 간결한 오류 메시지를 발생시키는 것이 우선입니다. Diffusers는 라이브러리를 가능한 한 쉽게 사용할 수 있도록 하는 것보다 사용자를 가르치는 것을 목표로 합니다.
- 복잡한 모델과 스케줄러 로직이 내부에서 마법처럼 처리하는 대신 노출됩니다. 스케줄러/샘플러는 서로에게 최소한의 종속성을 가지고 분리되어 있습니다. 이로써 사용자는 언롤된 노이즈 제거 루프를 작성해야 합니다. 그러나 이 분리는 디버깅을 더 쉽게하고 노이즈 제거 과정을 조정하거나 diffusers 모델이나 스케줄러를 교체하는 데 사용자에게 더 많은 제어권을 제공합니다.
- diffusers 파이프라인의 따로 훈련된 구성 요소인 text encoder, unet 및 variational autoencoder는 각각 자체 모델 클래스를 갖습니다. 이로써 사용자는 서로 다른 모델의 구성 요소 간의 상호 작용을 처리해야 하며, 직렬화 형식은 모델 구성 요소를 다른 파일로 분리합니다. 그러나 이는 디버깅과 커스터마이징을 더 쉽게합니다. DreamBooth나 Textual Inversion 훈련은 Diffusers의 'diffusion 파이프라인의 단일 구성 요소들을 분리할 수 있는 능력' 덕분에 매우 간단합니다.

## 추상화보다는 수정 가능하고 기여하기 쉬움을 [[tweakable-contributor-friendly-over-abstraction]]

라이브러리의 대부분에 대해 Diffusers는 [Transformers 라이브러리](https://github.com/huggingface/transformers)의 중요한 설계 원칙을 채택합니다, 바로 성급한 추상화보다는 copy-pasted 코드를 선호한다는 것입니다. 이 설계 원칙은 [Don't repeat yourself (DRY)](https://en.wikipedia.org/wiki/Don%27t_repeat_yourself)와 같은 인기 있는 설계 원칙과는 대조적으로 매우 의견이 분분한데요.
간단히 말해서, Transformers가 모델링 파일에 대해 수행하는 것처럼, Diffusers는 매우 낮은 수준의 추상화와 매우 독립적인 코드를 유지하는 것을 선호합니다. 함수, 긴 코드 블록, 심지어 클래스도 여러 파일에 복사할 수 있으며, 이는 처음에는 라이브러리를 유지할 수 없게 만드는 나쁜, 서투른 설계 선택으로 보일 수 있습니다. 하지만 이러한 설계는 매우 성공적이며, 커뮤니티 기반의 오픈 소스 기계 학습 라이브러리에 매우 적합합니다. 그 이유는 다음과 같습니다:
- 기계 학습은 패러다임, 모델 아키텍처 및 알고리즘이 빠르게 변화하는 매우 빠르게 움직이는 분야이기 때문에 오랜 기간 지속되는 코드 추상화를 정의하기가 매우 어렵습니다.
- 기계 학습 전문가들은 아이디어와 연구를 위해 기존 코드를 빠르게 조정할 수 있어야 하므로, 많은 추상화보다는 독립적인 코드를 선호합니다.
- 오픈 소스 라이브러리는 커뮤니티 기여에 의존하므로, 기여하기 쉬운 라이브러리를 구축해야 합니다. 코드가 추상화되면 의존성이 많아지고 읽기 어렵고 기여하기 어려워집니다. 기여자들은 중요한 기능을 망가뜨릴까 두려워하여 매우 추상화된 라이브러리에 기여하지 않게 됩니다. 라이브러리에 기여하는 것이 다른 기본 코드를 망가뜨릴 수 없다면, 잠재적인 새로운 기여자에게 더욱 환영받을 수 있을 뿐만 아니라 여러 부분에 대해 병렬적으로 검토하고 기여하기가 더 쉬워집니다.

Hugging Face에서는 이 설계를 **단일 파일 정책**이라고 부르며, 특정 클래스의 대부분의 코드가 단일하고 독립적인 파일에 작성되어야 한다는 의미입니다. 철학에 대해 자세히 알아보려면 [이 블로그 글](https://huggingface.co/blog/transformers-design-philosophy)을 참조할 수 있습니다.

Diffusers에서는 이러한 철학을 파이프라인과 스케줄러에 모두 따르지만, diffusion 모델에 대해서는 일부만 따릅니다. 일부만 따르는 이유는 Diffusion 파이프라인인 [DDPM](https://huggingface.co/docs/diffusers/api/pipelines/ddpm), [Stable Diffusion](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/overview#stable-diffusion-pipelines), [unCLIP (DALL·E 2)](https://huggingface.co/docs/diffusers/api/pipelines/unclip) 및 [Imagen](https://imagen.research.google/) 등 대부분의 diffusion 파이프라인은 동일한 diffusion 모델인 [UNet](https://huggingface.co/docs/diffusers/api/models/unet2d-cond)에 의존하기 때문입니다.

좋아요, 이제 🧨 Diffusers가 설계된 방식을 대략적으로 이해했을 것입니다 🤗.
우리는 이러한 설계 원칙을 일관되게 라이브러리 전체에 적용하려고 노력하고 있습니다. 그럼에도 불구하고 철학에 대한 일부 예외 사항이나 불행한 설계 선택이 있을 수 있습니다. 디자인에 대한 피드백이 있다면 [GitHub에서 직접](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feedback.md&title=) 알려주시면 감사하겠습니다.

## 디자인 철학 자세히 알아보기 [[design-philosophy-in-details]]

이제 디자인 철학의 세부 사항을 좀 더 자세히 살펴보겠습니다. Diffusers는 주로 세 가지 주요 클래스로 구성됩니다: [파이프라인](https://github.com/huggingface/diffusers/tree/main/src/diffusers/pipelines), [모델](https://github.com/huggingface/diffusers/tree/main/src/diffusers/models), 그리고 [스케줄러](https://github.com/huggingface/diffusers/tree/main/src/diffusers/schedulers). 각 클래스에 대한 더 자세한 설계 결정 사항을 살펴보겠습니다.

### 파이프라인 [[pipelines]]

파이프라인은 사용하기 쉽도록 설계되었으며 (따라서 [*쉬움보다는 간단함을*](#쉬움보다는-간단함을)을 100% 따르지는 않음), feature-complete하지 않으며, 추론을 위한 [모델](#모델)과 [스케줄러](#스케줄러)를 사용하는 방법의 예시로 간주될 수 있습니다.

다음과 같은 설계 원칙을 따릅니다:
- 파이프라인은 단일 파일 정책을 따릅니다. 모든 파이프라인은 src/diffusers/pipelines의 개별 디렉토리에 있습니다. 하나의 파이프라인 폴더는 하나의 diffusion 논문/프로젝트/릴리스에 해당합니다. 여러 파이프라인 파일은 하나의 파이프라인 폴더에 모을 수 있습니다. 예를 들어 [`src/diffusers/pipelines/stable-diffusion`](https://github.com/huggingface/diffusers/tree/main/src/diffusers/pipelines/stable_diffusion)에서 그렇게 하고 있습니다. 파이프라인이 유사한 기능을 공유하는 경우, [# Copied from mechanism](https://github.com/huggingface/diffusers/blob/125d783076e5bd9785beb05367a2d2566843a271/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py#L251)을 사용할 수 있습니다.
- 파이프라인은 모두 `DiffusionPipeline`을 상속합니다.
- 각 파이프라인은 서로 다른 모델 및 스케줄러 구성 요소로 구성되어 있으며, 이는 [`model_index.json` 파일](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5/blob/main/model_index.json)에 문서화되어 있으며, 파이프라인의 속성 이름과 동일한 이름으로 액세스할 수 있으며, [`DiffusionPipeline.components`](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.components) 함수를 통해 파이프라인 간에 공유할 수 있습니다.
- 각 파이프라인은 [`DiffusionPipeline.from_pretrained`](https://huggingface.co/docs/diffusers/main/en/api/diffusion_pipeline#diffusers.DiffusionPipeline.from_pretrained) 함수를 통해 로드할 수 있어야 합니다.
- 파이프라인은 추론에**만** 사용되어야 합니다.
- 파이프라인은 매우 가독성이 좋고, 이해하기 쉽고, 쉽게 조정할 수 있도록 설계되어야 합니다.
- 파이프라인은 서로 상호작용하고, 상위 수준 API에 쉽게 통합할 수 있도록 설계되어야 합니다.
- 파이프라인은 사용자 인터페이스가 feature-complete하지 않게 하는 것을 목표로 합니다. future-complete한 사용자 인터페이스를 원한다면 [InvokeAI](https://github.com/invoke-ai/InvokeAI), [Diffuzers](https://github.com/abhishekkrthakur/diffuzers), [lama-cleaner](https://github.com/Sanster/lama-cleaner)를 참조해야 합니다.
- 모든 파이프라인은 오로지 `__call__` 메소드를 통해 실행할 수 있어야 합니다. `__call__` 인자의 이름은 모든 파이프라인에서 공유되어야 합니다.
- 파이프라인은 해결하고자 하는 작업의 이름으로 지정되어야 합니다.
- 대부분의 경우에 새로운 diffusion 파이프라인은 새로운 파이프라인 폴더/파일에 구현되어야 합니다.

### 모델 [[models]]

모델은 [PyTorch의 Module 클래스](https://pytorch.org/docs/stable/generated/torch.nn.Module.html)의 자연스러운 확장이 되도록, 구성 가능한 툴박스로 설계되었습니다. 그리고 모델은 **단일 파일 정책**을 일부만 따릅니다.

다음과 같은 설계 원칙을 따릅니다:
- 모델은 **모델 아키텍처 유형**에 해당합니다. 예를 들어 `UNet2DConditionModel` 클래스는 2D 이미지 입력을 기대하고 일부 context에 의존하는 모든 UNet 변형들에 사용됩니다.
- 모든 모델은 [`src/diffusers/models`](https://github.com/huggingface/diffusers/tree/main/src/diffusers/models)에서 찾을 수 있으며, 각 모델 아키텍처는 해당 파일에 정의되어야 합니다. 예를 들어 [`unet_2d_condition.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_condition.py), [`transformer_2d.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/transformer_2d.py) 등이 있습니다.
- 모델은 **단일 파일 정책**을 따르지 않으며, [`attention.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention.py), [`resnet.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/resnet.py), [`embeddings.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/embeddings.py) 등과 같은 작은 모델 구성 요소를 사용해야 합니다. **참고**: 이는 Transformers의 모델링 파일과는 대조적으로 모델이 실제로 단일 파일 정책을 따르지 않음을 보여줍니다.
- 모델은 PyTorch의 `Module` 클래스와 마찬가지로 복잡성을 노출하고 명확한 오류 메시지를 제공해야 합니다.
- 모든 모델은 `ModelMixin`과 `ConfigMixin`을 상속합니다.
- 모델은 주요 코드 변경이 필요하지 않고, 역호환성을 유지하며, 메모리 또는 컴퓨팅과 관련한 중요한 이득을 제공할 때 성능을 위해 최적화할 수 있습니다.
- 모델은 기본적으로 가장 높은 정밀도와 가장 낮은 성능 설정을 가져야 합니다.
- Diffusers에 이미 있는 모델 아키텍처로 분류할 수 있는 새로운 모델 체크포인트를 통합할 때는 기존 모델 아키텍처를 새로운 체크포인트와 호환되도록 수정해야 합니다. 새로운 파일을 만들어야 하는 경우는 모델 아키텍처가 근본적으로 다른 경우에만 해당합니다.
- 모델은 미래의 변경 사항을 쉽게 확장할 수 있도록 설계되어야 합니다. 이는 공개 함수 인수들과 구성 인수들을 제한하고,미래의 변경 사항을 "예상"하는 것을 통해 달성할 수 있습니다. 예를 들어, 불리언 `is_..._type` 인수보다는 새로운 미래 유형에 쉽게 확장할 수 있는 문자열 "...type" 인수를 추가하는 것이 일반적으로 더 좋습니다. 새로운 모델 체크포인트가 작동하도록 하기 위해 기존 아키텍처에 최소한의 변경만을 가해야 합니다.
- 모델 디자인은 코드의 가독성과 간결성을 유지하는 것과 많은 모델 체크포인트를 지원하는 것 사이의 어려운 균형 조절입니다. 모델링 코드의 대부분은 새로운 모델 체크포인트를 위해 클래스를 수정하는 것이 좋지만, [UNet 블록](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_blocks.py) 및 [Attention 프로세서](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py)와 같이 코드를 장기적으로 간결하고 읽기 쉽게 유지하기 위해 새로운 클래스를 추가하는 예외도 있습니다.

### 스케줄러 [[schedulers]]

스케줄러는 추론을 위한 노이즈 제거 과정을 안내하고 훈련을 위한 노이즈 스케줄을 정의하는 역할을 합니다. 스케줄러는 개별 클래스로 설계되어 있으며, 로드 가능한 구성 파일과 **단일 파일 정책**을 엄격히 따릅니다.

다음과 같은 설계 원칙을 따릅니다:
- 모든 스케줄러는 [`src/diffusers/schedulers`](https://github.com/huggingface/diffusers/tree/main/src/diffusers/schedulers)에서 찾을 수 있습니다.
- 스케줄러는 큰 유틸리티 파일에서 가져오지 **않아야** 하며, 자체 포함성을 유지해야 합니다.
- 하나의 스케줄러 Python 파일은 하나의 스케줄러 알고리즘(논문에서 정의된 것과 같은)에 해당합니다.
- 스케줄러가 유사한 기능을 공유하는 경우, `# Copied from` 메커니즘을 사용할 수 있습니다.
- 모든 스케줄러는 `SchedulerMixin`과 `ConfigMixin`을 상속합니다.
- [`ConfigMixin.from_config`](https://huggingface.co/docs/diffusers/main/en/api/configuration#diffusers.ConfigMixin.from_config) 메소드를 사용하여 스케줄러를 쉽게 교체할 수 있습니다. 자세한 내용은 [여기](../using-diffusers/schedulers.md)에서 설명합니다.
- 모든 스케줄러는 `set_num_inference_steps`와 `step` 함수를 가져야 합니다. `set_num_inference_steps(...)`는 각 노이즈 제거 과정(즉, `step(...)`이 호출되기 전) 이전에 호출되어야 합니다.
- 각 스케줄러는 모델이 호출될 타임스텝의 배열인 `timesteps` 속성을 통해 루프를 돌 수 있는 타임스텝을 노출합니다.
- `step(...)` 함수는 예측된 모델 출력과 "현재" 샘플(x_t)을 입력으로 받고, "이전" 약간 더 노이즈가 제거된 샘플(x_t-1)을 반환합니다.
- 노이즈 제거 스케줄러의 복잡성을 고려하여, `step` 함수는 모든 복잡성을 노출하지 않으며, "블랙 박스"일 수 있습니다.
- 거의 모든 경우에 새로운 스케줄러는 새로운 스케줄링 파일에 구현되어야 합니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/conceptual/philosophy.md" />

### Diffusers에 기여하는 방법 🧨 [[how-to-contribute-to-diffusers-]]
https://huggingface.co/docs/diffusers/main/ko/conceptual/contribution.md

# Diffusers에 기여하는 방법 🧨 [[how-to-contribute-to-diffusers-]]

오픈 소스 커뮤니티에서의 기여를 환영합니다! 누구나 참여할 수 있으며, 코드뿐만 아니라 질문에 답변하거나 문서를 개선하는 등 모든 유형의 참여가 가치 있고 감사히 여겨집니다. 질문에 답변하고 다른 사람들을 도와주며 소통하고 문서를 개선하는 것은 모두 커뮤니티에게 큰 도움이 됩니다. 따라서 관심이 있다면 두려워하지 말고 참여해보세요!

누구나 우리의 공개 Discord 채널에서 👋 인사하며 시작할 수 있도록 장려합니다. 우리는 diffusion 모델의 최신 동향을 논의하고 질문을 하며 개인 프로젝트를 자랑하고 기여에 대해 서로 도와주거나 그냥 어울리기 위해 모이는 곳입니다☕. <a href="https://Discord.gg/G7tWnz98XR"><img alt="Join us on Discord" src="https://img.shields.io/discord/823813159592001537?color=5865F2&logo=discord&logoColor=white"></a>

어떤 방식으로든 기여하려는 경우, 우리는 개방적이고 환영하며 친근한 커뮤니티의 일부가 되기 위해 노력하고 있습니다. 우리의 [행동 강령](https://github.com/huggingface/diffusers/blob/main/CODE_OF_CONDUCT.md)을 읽고 상호 작용 중에 이를 존중하도록 주의해주시기 바랍니다. 또한 프로젝트를 안내하는 [윤리 지침](https://huggingface.co/docs/diffusers/conceptual/ethical_guidelines)에 익숙해지고 동일한 투명성과 책임성의 원칙을 준수해주시기를 부탁드립니다.

우리는 커뮤니티로부터의 피드백을 매우 중요하게 생각하므로, 라이브러리를 개선하는 데 도움이 될 가치 있는 피드백이 있다고 생각되면 망설이지 말고 의견을 제시해주세요 - 모든 메시지, 댓글, 이슈, Pull Request(PR)는 읽히고 고려됩니다.

## 개요 [[overview]]

이슈에 있는 질문에 답변하는 것에서부터 코어 라이브러리에 새로운 diffusion 모델을 추가하는 것까지 다양한 방법으로 기여를 할 수 있습니다.

이어지는 부분에서 우리는 다양한 방법의 기여에 대한 개요를 난이도에 따라 오름차순으로 정리하였습니다. 모든 기여는 커뮤니티에게 가치가 있습니다.

1. [Diffusers 토론 포럼](https://discuss.huggingface.co/c/discussion-related-to-httpsgithubcomhuggingfacediffusers)이나 [Discord](https://discord.gg/G7tWnz98XR)에서 질문에 대답하거나 질문을 할 수 있습니다.
2. [GitHub Issues 탭](https://github.com/huggingface/diffusers/issues/new/choose)에서 새로운 이슈를 열 수 있습니다.
3. [GitHub Issues 탭](https://github.com/huggingface/diffusers/issues)에서 이슈에 대답할 수 있습니다.
4. "Good first issue" 라벨이 지정된 간단한 이슈를 수정할 수 있습니다. [여기](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22)를 참조하세요.
5. [문서](https://github.com/huggingface/diffusers/tree/main/docs/source)에 기여할 수 있습니다.
6. [Community Pipeline](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3Acommunity-examples)에 기여할 수 있습니다.
7. [예제](https://github.com/huggingface/diffusers/tree/main/examples)에 기여할 수 있습니다.
8. "Good second issue" 라벨이 지정된 어려운 이슈를 수정할 수 있습니다. [여기](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22Good+second+issue%22)를 참조하세요.
9. 새로운 파이프라인, 모델 또는 스케줄러를 추가할 수 있습니다. ["새로운 파이프라인/모델"](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22New+pipeline%2Fmodel%22) 및 ["새로운 스케줄러"](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22New+scheduler%22) 이슈를 참조하세요. 이 기여에 대해서는 [디자인 철학](https://github.com/huggingface/diffusers/blob/main/PHILOSOPHY.md)을 확인해주세요.

앞서 말한 대로, **모든 기여는 커뮤니티에게 가치가 있습니다**. 이어지는 부분에서 각 기여에 대해 조금 더 자세히 설명하겠습니다.

4부터 9까지의 모든 기여에는 Pull Request을 열어야 합니다. [Pull Request 열기](#how-to-open-a-pr)에서 자세히 설명되어 있습니다.

### 1. Diffusers 토론 포럼이나 Diffusers Discord에서 질문하고 답변하기 [[1-asking-and-answering-questions-on-the-diffusers-discussion-forum-or-on-the-diffusers-discord]]

Diffusers 라이브러리와 관련된 모든 질문이나 의견은 [토론 포럼](https://discuss.huggingface.co/c/discussion-related-to-httpsgithubcomhuggingfacediffusers/63)이나 [Discord](https://discord.gg/G7tWnz98XR)에서 할 수 있습니다. 이러한 질문과 의견에는 다음과 같은 내용이 포함됩니다(하지만 이에 국한되지는 않습니다):
- 지식을 공유하기 위해서 훈련 또는 추론 실험에 대한 결과 보고
- 개인 프로젝트 소개
- 비공식 훈련 예제에 대한 질문
- 프로젝트 제안
- 일반적인 피드백
- 논문 요약
- Diffusers 라이브러리를 기반으로 하는 개인 프로젝트에 대한 도움 요청
- 일반적인 질문
- Diffusion 모델에 대한 윤리적 질문
- ...

포럼이나 Discord에서 질문을 하면 커뮤니티가 지식을 공개적으로 공유하도록 장려되며, 향후 동일한 질문을 가진 초보자에게도 도움이 될 수 있습니다. 따라서 궁금한 질문은 언제든지 하시기 바랍니다.
또한, 이러한 질문에 답변하는 것은 커뮤니티에게 매우 큰 도움이 됩니다. 왜냐하면 이렇게 하면 모두가 학습할 수 있는 공개적인 지식을 문서화하기 때문입니다.

**주의**하십시오. 질문이나 답변에 투자하는 노력이 많을수록 공개적으로 문서화된 지식의 품질이 높아집니다. 마찬가지로, 잘 정의되고 잘 답변된 질문은 모두에게 접근 가능한 고품질 지식 데이터베이스를 만들어줍니다. 반면에 잘못된 질문이나 답변은 공개 지식 데이터베이스의 전반적인 품질을 낮출 수 있습니다.
간단히 말해서, 고품질의 질문이나 답변은 *명확하고 간결하며 관련성이 있으며 이해하기 쉽고 접근 가능하며 잘 형식화되어 있어야* 합니다. 자세한 내용은 [좋은 이슈 작성 방법](#how-to-write-a-good-issue) 섹션을 참조하십시오.

**채널에 대한 참고사항**:
[*포럼*](https://discuss.huggingface.co/c/discussion-related-to-httpsgithubcomhuggingfacediffusers/63)은 구글과 같은 검색 엔진에서 더 잘 색인화됩니다. 게시물은 인기에 따라 순위가 매겨지며, 시간순으로 정렬되지 않습니다. 따라서 이전에 게시한 질문과 답변을 쉽게 찾을 수 있습니다.
또한, 포럼에 게시된 질문과 답변은 쉽게 링크할 수 있습니다.
반면 *Discord*는 채팅 형식으로 되어 있어 빠른 대화를 유도합니다.
질문에 대한 답변을 빠르게 받을 수는 있겠지만, 시간이 지나면 질문이 더 이상 보이지 않습니다. 또한, Discord에서 이전에 게시된 정보를 찾는 것은 훨씬 어렵습니다. 따라서 포럼을 사용하여 고품질의 질문과 답변을 하여 커뮤니티를 위한 오래 지속되는 지식을 만들기를 권장합니다. Discord에서의 토론이 매우 흥미로운 답변과 결론을 이끌어내는 경우, 해당 정보를 포럼에 게시하여 향후 독자들에게 더 쉽게 액세스할 수 있도록 권장합니다.

### 2. GitHub 이슈 탭에서 새로운 이슈 열기 [[2-opening-new-issues-on-the-github-issues-tab]]

🧨 Diffusers 라이브러리는 사용자들이 마주치는 문제를 알려주는 덕분에 견고하고 신뢰할 수 있습니다. 따라서 이슈를 보고해주셔서 감사합니다.

기억해주세요, GitHub 이슈는 Diffusers 라이브러리와 직접적으로 관련된 기술적인 질문, 버그 리포트, 기능 요청 또는 라이브러리 디자인에 대한 피드백에 사용됩니다.

간단히 말해서, Diffusers 라이브러리의 **코드와 관련되지 않은** 모든 것(문서 포함)은 GitHub가 아닌 [포럼](https://discuss.huggingface.co/c/discussion-related-to-httpsgithubcomhuggingfacediffusers/63)이나 [Discord](https://discord.gg/G7tWnz98XR)에서 질문해야 합니다.

**새로운 이슈를 열 때 다음 가이드라인을 고려해주세요**:
- 이미 같은 이슈가 있는지 검색했는지 확인해주세요(GitHub의 이슈 탭에서 검색 기능을 사용하세요).
- 다른(관련된) 이슈에 새로운 이슈를 보고하지 말아주세요. 다른 이슈와 관련이 높다면, 새로운 이슈를 열고 관련 이슈에 링크를 걸어주세요.
- 이슈를 영어로 작성해주세요. 영어에 익숙하지 않다면, [DeepL](https://www.deepl.com/translator)과 같은 뛰어난 무료 온라인 번역 서비스를 사용하여 모국어에서 영어로 번역해주세요.
- 이슈가 최신 Diffusers 버전으로 업데이트하면 해결될 수 있는지 확인해주세요. 이슈를 게시하기 전에 `python -c "import diffusers; print(diffusers.__version__)"` 명령을 실행하여 현재 사용 중인 Diffusers 버전이 최신 버전과 일치하거나 더 높은지 확인해주세요.
- 새로운 이슈를 열 때 투자하는 노력이 많을수록 답변의 품질이 높아지고 Diffusers 이슈 전체의 품질도 향상됩니다.

#### 2.1 재현 가능한 최소한의 버그 리포트 [[21-reproducible-minimal-bug-reports]]


버그 리포트는 항상 재현 가능한 코드 조각을 포함하고 가능한 한 최소한이어야 하며 간결해야 합니다.
자세히 말하면:
- 버그를 가능한 한 좁혀야 합니다. **전체 코드 파일을 그냥 던지지 마세요**.
- 코드의 서식을 지정해야 합니다.
- Diffusers가 의존하는 외부 라이브러리를 제외한 다른 외부 라이브러리는 포함하지 마십시오.
- **항상** 사용자 환경에 대한 모든 필요한 정보를 제공하세요. 이를 위해 쉘에서 `diffusers-cli env`를 실행하고 표시된 정보를 이슈에 복사하여 붙여넣을 수 있습니다.
- 이슈를 설명해야 합니다. 독자가 문제가 무엇인지, 왜 문제가 되는지 모른다면 이슈를 해결할 수 없습니다. 
- **항상** 독자가 가능한 한 적은 노력으로 문제를 재현할 수 있어야 합니다. 코드 조각이 라이브러리가 없거나 정의되지 않은 변수 때문에 실행되지 않는 경우 독자가 도움을 줄 수 없습니다. 재현 가능한 코드 조각이 가능한 한 최소화되고 간단한 Python 셸에 복사하여 붙여넣을 수 있도록 해야 합니다.
- 문제를 재현하기 위해 모델과/또는 데이터셋이 필요한 경우 독자가 해당 모델이나 데이터셋에 접근할 수 있도록 해야 합니다. 모델이나 데이터셋을 [Hub](https://huggingface.co)에 업로드하여 쉽게 다운로드할 수 있도록 할 수 있습니다. 문제 재현을 가능한 한 쉽게하기 위해 모델과 데이터셋을 가능한 한 작게 유지하려고 노력하세요.

자세한 내용은 [좋은 이슈 작성 방법](#how-to-write-a-good-issue) 섹션을 참조하세요.

버그 리포트를 열려면 [여기](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=bug&projects=&template=bug-report.yml)를 클릭하세요.


#### 2.2. 기능 요청 [[22-feature-requests]]

세계적인 기능 요청은 다음 사항을 다룹니다:

1. 먼저 동기부여:
* 라이브러리와 관련된 문제/불만이 있나요? 그렇다면 왜 그런지 설명해주세요. 문제를 보여주는 코드 조각을 제공하는 것이 가장 좋습니다.
* 프로젝트에 필요한 기능인가요? 우리는 그에 대해 듣고 싶습니다!
* 커뮤니티에 도움이 될 수 있는 것을 작업했고 그것에 대해 생각하고 있는가요? 멋지네요! 어떤 문제를 해결했는지 알려주세요.
2. 기능을 *상세히 설명하는* 문단을 작성해주세요;
3. 향후 사용을 보여주는 **코드 조각**을 제공해주세요;
4. 논문과 관련된 내용인 경우 링크를 첨부해주세요;
5. 도움이 될 수 있다고 생각되는 추가 정보(그림, 스크린샷 등)를 첨부해주세요.

기능 요청은 [여기](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feature_request.md&title=)에서 열 수 있습니다.

#### 2.3 피드백 [[23-feedback]]

라이브러리 디자인과 그것이 왜 좋은지 또는 나쁜지에 대한 이유에 대한 피드백은 핵심 메인테이너가 사용자 친화적인 라이브러리를 만드는 데 엄청난 도움이 됩니다. 현재 디자인 철학을 이해하려면 [여기](https://huggingface.co/docs/diffusers/conceptual/philosophy)를 참조해 주세요. 특정 디자인 선택이 현재 디자인 철학과 맞지 않는다고 생각되면, 그 이유와 어떻게 변경되어야 하는지 설명해 주세요. 반대로 특정 디자인 선택이 디자인 철학을 너무 따르기 때문에 사용 사례를 제한한다고 생각되면, 그 이유와 어떻게 변경되어야 하는지 설명해 주세요. 특정 디자인 선택이 매우 유용하다고 생각되면, 향후 디자인 결정에 큰 도움이 되므로 이에 대한 의견을 남겨 주세요.

피드백에 관한 이슈는 [여기](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feedback.md&title=)에서 열 수 있습니다.

#### 2.4 기술적인 질문 [[24-technical-questions]]

기술적인 질문은 주로 라이브러리의 특정 코드가 왜 특정 방식으로 작성되었는지 또는 코드의 특정 부분이 무엇을 하는지에 대한 질문입니다. 질문하신 코드 부분에 대한 링크를 제공하고 해당 코드 부분이 이해하기 어려운 이유에 대한 자세한 설명을 해주시기 바랍니다.

기술적인 질문에 관한 이슈를 [여기](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=bug&template=bug-report.yml)에서 열 수 있습니다.

#### 2.5 새로운 모델, 스케줄러 또는 파이프라인 추가 제안 [[25-proposal-to-add-a-new-model-scheduler-or-pipeline]]

만약 diffusion 모델 커뮤니티에서 Diffusers 라이브러리에 추가하고 싶은 새로운 모델, 파이프라인 또는 스케줄러가 있다면, 다음 정보를 제공해주세요:

* Diffusion 파이프라인, 모델 또는 스케줄러에 대한 간단한 설명과 논문 또는 공개된 버전의 링크
* 해당 모델의 오픈 소스 구현에 대한 링크
* 모델 가중치가 있는 경우, 가중치의 링크

직접 모델에 기여하고 싶다면, 가장 잘 안내해드릴 수 있습니다. 또한, 가능하다면 구성 요소(모델, 스케줄러, 파이프라인 등)의 원저자를 GitHub 핸들로 태그하는 것을 잊지 마세요.

모델/파이프라인/스케줄러에 대한 요청을 [여기](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=New+model%2Fpipeline%2Fscheduler&template=new-model-addition.yml)에서 열 수 있습니다.

### 3. GitHub 이슈 탭에서 문제에 대한 답변하기 [[3-answering-issues-on-the-github-issues-tab]]

GitHub에서 이슈에 대한 답변을 하기 위해서는 Diffusers에 대한 기술적인 지식이 필요할 수 있지만, 정확한 답변이 아니더라도 모두가 시도해기를 권장합니다. 이슈에 대한 고품질 답변을 제공하기 위한 몇 가지 팁:
- 가능한 한 간결하고 최소한으로 유지합니다.
- 주제에 집중합니다. 이슈에 대한 답변은 해당 이슈에 관련된 내용에만 집중해야 합니다.
- 자신의 주장을 증명하거나 장려하는 코드, 논문 또는 기타 출처는 링크를 제공하세요.
- 코드로 답변합니다. 간단한 코드 조각이 이슈에 대한 답변이거나 이슈를 해결하는 방법을 보여준다면, 완전히 재현 가능한 코드 조각을 제공해주세요.

또한, 많은 이슈들은 단순히 주제와 무관하거나 다른 이슈의 중복이거나 관련이 없는 경우가 많습니다. 이러한 이슈들에 대한 답변을 제공하고, 이슈 작성자에게 더 정확한 정보를 제공하거나, 중복된 이슈에 대한 링크를 제공하거나, [포럼](https://discuss.huggingface.co/c/discussion-related-to-httpsgithubcomhuggingfacediffusers/63) 이나 [Discord](https://discord.gg/G7tWnz98XR)로 리디렉션하는 것은 메인테이너에게 큰 도움이 됩니다.

이슈가 올바른 버그 보고서이고 소스 코드에서 수정이 필요하다고 확인한 경우, 다음 섹션을 살펴보세요.

다음 모든 기여에 대해서는 PR을 열여야 합니다. [Pull Request 열기](#how-to-open-a-pr) 섹션에서 자세히 설명되어 있습니다.

### 4. "Good first issue" 고치기 [[4-fixing-a-good-first-issue]]

*Good first issues*는 [Good first issue](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22) 라벨로 표시됩니다. 일반적으로, 이슈는 이미 잠재적인 해결책이 어떻게 보이는지 설명하고 있어서 수정하기 쉽습니다.
만약 이슈가 아직 닫히지 않았고 이 문제를 해결해보고 싶다면, "이 이슈를 해결해보고 싶습니다."라는 메시지를 남기면 됩니다. 일반적으로 세 가지 시나리오가 있습니다:
- a.) 이슈 설명에 이미 수정 사항을 제안하는 경우, 해결책이 이해되고 합리적으로 보인다면, PR 또는 드래프트 PR을 열어서 수정할 수 있습니다.
- b.) 이슈 설명에 수정 사항이 제안되어 있지 않은 경우, 제안한 수정 사항이 가능할지 물어볼 수 있고, Diffusers 팀의 누군가가 곧 답변해줄 것입니다. 만약 어떻게 수정할지 좋은 아이디어가 있다면, 직접 PR을 열어도 됩니다.
- c.) 이미 이 문제를 해결하기 위해 열린 PR이 있지만, 이슈가 아직 닫히지 않았습니다. PR이 더 이상 진행되지 않았다면, 새로운 PR을 열고 이전 PR에 링크를 걸면 됩니다. PR은 종종 원래 기여자가 갑자기 시간을 내지 못해 더 이상 진행하지 못하는 경우에 더 이상 진행되지 않게 됩니다. 이는 오픈 소스에서 자주 발생하는 일이며 매우 정상적인 상황입니다. 이 경우, 커뮤니티는 새로 시도하고 기존 PR의 지식을 활용해주면 매우 기쁠 것입니다. 이미 PR이 있고 활성화되어 있다면, 제안을 해주거나 PR을 검토하거나 PR에 기여할 수 있는지 물어보는 등 작성자를 도와줄 수 있습니다.


### 5. 문서에 기여하기 [[5-contribute-to-the-documentation]]

좋은 라이브러리는 항상 좋은 문서를 갖고 있습니다! 공식 문서는 라이브러리를 처음 사용하는 사용자들에게 첫 번째 접점 중 하나이며, 따라서 문서에 기여하는 것은 매우 가치 있는 기여입니다.

라이브러리에 기여하는 방법은 다양합니다:

- 맞춤법이나 문법 오류를 수정합니다.
- 공식 문서가 이상하게 표시되거나 링크가 깨진 경우, 올바르게 수정하는 데 시간을 내주시면 매우 기쁠 것입니다.
- 문서의 입력 또는 출력 텐서의 모양이나 차원을 수정합니다.
- 이해하기 어렵거나 잘못된 문서를 명확하게 합니다.
- 오래된 코드 예제를 업데이트합니다.
- 문서를 다른 언어로 번역합니다.

[공식 Diffusers 문서 페이지](https://huggingface.co/docs/diffusers/index)에 표시된 모든 내용은 공식 문서의 일부이며, 해당 [문서 소스](https://github.com/huggingface/diffusers/tree/main/docs/source)에서 수정할 수 있습니다.

문서에 대한 변경 사항을 로컬에서 확인하는 방법은 [이 페이지](https://github.com/huggingface/diffusers/tree/main/docs)를 참조해주세요.


### 6. 커뮤니티 파이프라인에 기여하기 [[6-contribute-a-community-pipeline]]

> [!TIP]
> 커뮤니티 파이프라인에 대해 자세히 알아보려면 [커뮤니티 파이프라인](../using-diffusers/custom_pipeline_overview#community-pipelines) 가이드를 읽어보세요. 커뮤니티 파이프라인이 왜 필요한지 궁금하다면 GitHub 이슈 [#841](https://github.com/huggingface/diffusers/issues/841)를 확인해보세요 (기본적으로, 우리는 diffusion 모델이 추론에 사용될 수 있는 모든 방법을 유지할 수 없지만 커뮤니티가 이를 구축하는 것을 방해하고 싶지 않습니다).

커뮤니티 파이프라인에 기여하는 것은 창의성과 작업을 커뮤니티와 공유하는 좋은 방법입니다. `DiffusionPipeline`을 기반으로 빌드하여 `custom_pipeline` 매개변수를 설정함으로써 누구나 로드하고 사용할 수 있도록 할 수 있습니다. 이 섹션에서는 UNet이 단일 순방향 패스만 수행하고 스케줄러를 한 번 호출하는 간단한 파이프라인 (단계별 파이프라인)을 만드는 방법을 안내합니다.

1. 커뮤니티 파이프라인을 위한 one_step_unet.py 파일을 생성하세요. 이 파일은 사용자에 의해 설치되는 패키지를 포함할 수 있지만, `DiffusionPipeline`에서 모델 가중치와 스케줄러 구성을 로드하기 위해 하나의 파이프라인 클래스만 있어야 합니다. `__init__` 함수에 UNet과 스케줄러를 추가하세요.

    또한 `save_pretrained()`를 사용하여 파이프라인과 그 구성 요소를 저장할 수 있도록 `register_modules` 함수를 추가해야 합니다.

```py
from diffusers import DiffusionPipeline
import torch

class UnetSchedulerOneForwardPipeline(DiffusionPipeline):
    def __init__(self, unet, scheduler):
        super().__init__()

        self.register_modules(unet=unet, scheduler=scheduler)
```

1. forward 패스에서 (`__call__`로 정의하는 것을 추천합니다), 원하는 어떤 기능이든 추가할 수 있습니다. "one-step" 파이프라인의 경우, 무작위 이미지를 생성하고 `timestep=1`로 설정하여 UNet과 스케줄러를 한 번 호출합니다.

```py
  from diffusers import DiffusionPipeline
  import torch

  class UnetSchedulerOneForwardPipeline(DiffusionPipeline):
      def __init__(self, unet, scheduler):
          super().__init__()

          self.register_modules(unet=unet, scheduler=scheduler)

      def __call__(self):
          image = torch.randn(
              (1, self.unet.config.in_channels, self.unet.config.sample_size, self.unet.config.sample_size),
          )
          timestep = 1

          model_output = self.unet(image, timestep).sample
          scheduler_output = self.scheduler.step(model_output, timestep, image).prev_sample

          return scheduler_output
```

이제 UNet과 스케줄러를 전달하여 파이프라인을 실행하거나, 파이프라인 구조가 동일한 경우 사전 학습된 가중치를 로드할 수 있습니다.

```py
from diffusers import DDPMScheduler, UNet2DModel

scheduler = DDPMScheduler()
unet = UNet2DModel()

pipeline = UnetSchedulerOneForwardPipeline(unet=unet, scheduler=scheduler)
output = pipeline()
# load pretrained weights
pipeline = UnetSchedulerOneForwardPipeline.from_pretrained("google/ddpm-cifar10-32", use_safetensors=True)
output = pipeline()
```

파이프라인을 GitHub 커뮤니티 파이프라인 또는 Hub 커뮤니티 파이프라인으로 공유할 수 있습니다.

<hfoptions id="pipeline type">
<hfoption id="GitHub pipeline">

GitHub 파이프라인을 공유하려면 Diffusers [저장소](https://github.com/huggingface/diffusers)에서 Pull Request를 열고 one_step_unet.py 파일을 [examples/community](https://github.com/huggingface/diffusers/tree/main/examples/community) 하위 폴더에 추가하세요.

</hfoption>
<hfoption id="Hub pipeline">

Hub 파이프라인을 공유하려면, 허브에 모델 저장소를 생성하고 one_step_unet.py 파일을 업로드하세요.

</hfoption>
</hfoptions>

### 7. 훈련 예제에 기여하기 [[7-contribute-to-training-examples]]

Diffusers 예제는 [examples](https://github.com/huggingface/diffusers/tree/main/examples) 폴더에 있는 훈련 스크립트의 모음입니다.

두 가지 유형의 훈련 예제를 지원합니다:

- 공식 훈련 예제
- 연구용 훈련 예제

연구용 훈련 예제는 [examples/research_projects](https://github.com/huggingface/diffusers/tree/main/examples/research_projects)에 위치하며, 공식 훈련 예제는 `research_projects` 및 `community` 폴더를 제외한 [examples](https://github.com/huggingface/diffusers/tree/main/examples)의 모든 폴더를 포함합니다.
공식 훈련 예제는 Diffusers의 핵심 메인테이너가 유지 관리하며, 연구용 훈련 예제는 커뮤니티가 유지 관리합니다.
이는 공식 파이프라인 vs 커뮤니티 파이프라인에 대한 [6. 커뮤니티 파이프라인 기여하기](#6-contribute-a-community-pipeline)에서 제시한 이유와 동일합니다: 핵심 메인테이너가 diffusion 모델의 모든 가능한 훈련 방법을 유지 관리하는 것은 현실적으로 불가능합니다.
Diffusers 핵심 메인테이너와 커뮤니티가 특정 훈련 패러다임을 너무 실험적이거나 충분히 대중적이지 않다고 판단한다면, 해당 훈련 코드는 `research_projects` 폴더에 넣고 작성자에 의해 관리되어야 합니다.

공식 훈련 및 연구 예제는 하나 이상의 훈련 스크립트, requirements.txt 파일 및 README.md 파일을 포함하는 디렉토리로 구성됩니다. 사용자가 훈련 예제를 사용하려면 리포지토리를 복제해야 합니다:

```bash
git clone https://github.com/huggingface/diffusers
```

그리고 훈련에 필요한 모든 추가적인 의존성도 설치해야 합니다:

```bash
pip install -r /examples/<your-example-folder>/requirements.txt
```

따라서 예제를 추가할 때, `requirements.txt` 파일은 훈련 예제에 필요한 모든 pip 종속성을 정의해야 합니다. 이렇게 설치된 모든 종속성을 사용하여 사용자가 예제의 훈련 스크립트를 실행할 수 있어야 합니다. 예를 들어, [DreamBooth `requirements.txt` 파일](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/requirements.txt)을 참조하세요.

Diffusers 라이브러리의 훈련 예제는 다음 철학을 따라야 합니다:
- 예제를 실행하는 데 필요한 모든 코드는 하나의 Python 파일에 있어야 합니다.
- 사용자는 명령 줄에서 `python <your-example>.py --args`와 같이 예제를 실행할 수 있어야 합니다.
- 예제는 간단하게 유지되어야 하며, Diffusers를 사용한 훈련 방법을 보여주는 **예시**로 사용되어야 합니다. 예제 스크립트의 목적은 최첨단 diffusion 모델을 만드는 것이 아니라, 너무 많은 사용자 정의 로직을 추가하지 않고 이미 알려진 훈련 방법을 재현하는 것입니다. 이 점의 부산물로서, 예제는 좋은 교육 자료로써의 역할을 하기 위해 노력합니다.

예제에 기여하기 위해서는, 이미 존재하는 예제인 [dreambooth](https://github.com/huggingface/diffusers/blob/main/examples/dreambooth/train_dreambooth.py)와 같은 예제를 참고하여 어떻게 보여야 하는지에 대한 아이디어를 얻는 것이 매우 권장됩니다.
Diffusers와 긴밀하게 통합되어 있기 때문에, 기여자들이 [Accelerate 라이브러리](https://github.com/huggingface/accelerate)를 사용하는 것을 강력히 권장합니다.
예제 스크립트가 작동하는 경우, 반드시 예제를 정확하게 사용하는 방법을 설명하는 포괄적인 `README.md`를 추가해야 합니다. 이 README에는 다음이 포함되어야 합니다:
- [여기](https://github.com/huggingface/diffusers/tree/main/examples/dreambooth#running-locally-with-pytorch)에 표시된 예제 스크립트를 실행하는 방법에 대한 예제 명령어.
- [여기](https://api.wandb.ai/report/patrickvonplaten/xm6cd5q5)에 표시된 훈련 결과 (로그, 모델 등)에 대한 링크로 사용자가 기대할 수 있는 내용을 보여줍니다.
- 비공식/연구용 훈련 예제를 추가하는 경우, **반드시** git 핸들을 포함하여 이 훈련 예제를 유지 관리할 것임을 명시하는 문장을 추가해야 합니다. [여기](https://github.com/huggingface/diffusers/tree/main/examples/research_projects/intel_opts#diffusers-examples-with-intel-optimizations)에 표시된 것과 같습니다.

만약 공식 훈련 예제에 기여하는 경우, [examples/test_examples.py](https://github.com/huggingface/diffusers/blob/main/examples/test_examples.py)에 테스트를 추가하는 것도 확인해주세요. 비공식 훈련 예제에는 이 작업이 필요하지 않습니다.

### 8. "Good second issue" 고치기 [[8-fixing-a-good-second-issue]]

"Good second issue"는 [Good second issue](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22Good+second+issue%22) 라벨로 표시됩니다. Good second issue는 [Good first issues](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22good+first+issue%22)보다 해결하기가 더 복잡합니다.
이슈 설명은 일반적으로 이슈를 해결하는 방법에 대해 덜 구체적이며, 관심 있는 기여자는 라이브러리에 대한 꽤 깊은 이해가 필요합니다.
Good second issue를 해결하고자 하는 경우, 해당 이슈를 해결하기 위해 PR을 열고 PR을 이슈에 링크하세요. 이미 해당 이슈에 대한 PR이 열려있지만 병합되지 않은 경우, 왜 병합되지 않았는지 이해하기 위해 살펴보고 개선된 PR을 열어보세요.
Good second issue는 일반적으로 Good first issue 이슈보다 병합하기가 더 어려우므로, 핵심 메인테이너에게 도움을 요청하는 것이 좋습니다. PR이 거의 완료된 경우, 핵심 메인테이너는 PR에 참여하여 커밋하고 병합을 진행할 수 있습니다.

### 9. 파이프라인, 모델, 스케줄러 추가하기 [[9-adding-pipelines-models-schedulers]]

파이프라인, 모델, 스케줄러는 Diffusers 라이브러리에서 가장 중요한 부분입니다.
이들은 최첨단 diffusion 기술에 쉽게 접근하도록 하며, 따라서 커뮤니티가 강력한 생성형 AI 애플리케이션을 만들 수 있도록 합니다.

새로운 모델, 파이프라인 또는 스케줄러를 추가함으로써, 사용자 인터페이스에 새로운 강력한 사용 사례를 활성화할 수 있으며, 이는 전체 생성형 AI 생태계에 매우 중요한 가치를 제공할 수 있습니다.

Diffusers에는 세 가지 구성 요소에 대한 여러 개발 요청이 있습니다. 특정 구성 요소를 아직 정확히 어떤 것을 추가하고 싶은지 모르는 경우, 다음 링크를 참조하세요:
- [모델 또는 파이프라인](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22New+pipeline%2Fmodel%22)
- [스케줄러](https://github.com/huggingface/diffusers/issues?q=is%3Aopen+is%3Aissue+label%3A%22New+scheduler%22)


세 가지 구성 요소를 추가하기 전에, [철학 가이드](philosophy)를 읽어보는 것을 강력히 권장합니다. 세 가지 구성 요소 중 어느 것을 추가하든, 디자인 철학과 관련된 API 일관성을 유지하기 위해 우리의 디자인 철학과 크게 다른 구성 요소는 병합할 수 없습니다. 디자인 선택에 근본적으로 동의하지 않는 경우, [피드백 이슈](https://github.com/huggingface/diffusers/issues/new?assignees=&labels=&template=feedback.md&title=)를 열어 해당 디자인 패턴/선택이 라이브러리 전체에서 변경되어야 하는지, 디자인 철학을 업데이트해야 하는지에 대해 논의할 수 있습니다. 라이브러리 전체의 일관성은 우리에게 매우 중요합니다.

PR에 원본 코드베이스/논문 링크를 추가하고, 가능하면 PR에서 원래 작성자에게 직접 알림을 보내어 진행 상황을 따라갈 수 있도록 해주세요.

PR에서 막힌 경우나 도움이 필요한 경우, 첫 번째 리뷰나 도움을 요청하는 메시지를 남기는 것을 주저하지 마세요.

#### Copied from mechanism [[copied-from-mechanism]]

`# Copied from mechanism` 은 파이프라인, 모델 또는 스케줄러 코드를 추가할 때 이해해야 할 독특하고 중요한 기능입니다. 이것은 Diffusers 코드베이스 전반에서 볼 수 있으며, 이를 사용하는 이유는 코드베이스를 이해하고 유지 관리하기 쉽게 만들기 위해서입니다. `# Copied from mechanism` 으로 표시된 코드는 복사한 코드와 정확히 동일하도록 강제됩니다. 이렇게 하면 `make fix-copies`를 실행할 때마다 여러 파일에 걸쳐 변경 사항을 쉽게 업데이트하고 전파할 수 있습니다.

예를 들어, 아래 코드 예제에서 `StableDiffusionPipelineOutput`은 원래 코드이며, `AltDiffusionPipelineOutput`은 `# Copied from mechanism`을 사용하여 복사합니다. 유일한 차이점은 클래스 접두사를 `Stable`에서 `Alt`로 변경한 것입니다.

```py
# Copied from diffusers.pipelines.stable_diffusion.pipeline_output.StableDiffusionPipelineOutput with Stable->Alt
class AltDiffusionPipelineOutput(BaseOutput):
    """
    Output class for Alt Diffusion pipelines.

    Args:
        images (`List[PIL.Image.Image]` or `np.ndarray`)
            List of denoised PIL images of length `batch_size` or NumPy array of shape `(batch_size, height, width,
            num_channels)`.
        nsfw_content_detected (`List[bool]`)
            List indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content or
            `None` if safety checking could not be performed.
    """
```

더 자세히 알고 싶다면 [~Don't~ Repeat Yourself*](https://huggingface.co/blog/transformers-design-philosophy#4-machine-learning-models-are-static) 블로그 포스트의 이 섹션을 읽어보세요.

## 좋은 이슈 작성 방법 [[how-to-write-a-good-issue]]

**이슈를 잘 작성할수록 빠르게 해결될 가능성이 높아집니다.**

1. 이슈에 적절한 템플릿을 사용했는지 확인하세요. [새 이슈를 열 때](https://github.com/huggingface/diffusers/issues/new/choose) 올바른 템플릿을 선택해야 합니다. *버그 보고서*, *기능 요청*, *API 디자인에 대한 피드백*, *새로운 모델/파이프라인/스케줄러 추가*, *포럼*, 또는 빈 이슈 중에서 선택하세요. 이슈를 열 때 올바른 템플릿을 선택하는 것이 중요합니다.
2. **명확성**: 이슈에 적합한 제목을 지정하세요. 이슈 설명을 가능한 간단하게 작성하세요. 이슈를 이해하고 해결하는 데 걸리는 시간을 줄이기 위해 가능한 한 명확하게 작성하세요. 하나의 이슈에 대해 여러 문제를 포함하지 않도록 주의하세요. 여러 문제를 발견한 경우, 각각의 이슈를 개별적으로 열어주세요. 버그인 경우, 어떤 버그인지 가능한 한 정확하게 설명해야 합니다. "diffusers에서 오류"와 같이 간단히 작성하지 마세요.
3. **재현 가능성**: 재현 가능한 코드 조각이 없으면 해결할 수 없습니다. 버그를 발견한 경우, 유지 관리자는 그 버그를 재현할 수 있어야 합니다. 이슈에 재현 가능한 코드 조각을 포함해야 합니다. 코드 조각은 Python 인터프리터에 복사하여 붙여넣을 수 있는 형태여야 합니다. 코드 조각이 작동해야 합니다. 즉, 누락된 import나 이미지에 대한 링크가 없어야 합니다. 이슈에는 오류 메시지와 정확히 동일한 오류 메시지를 재현하기 위해 수정하지 않고 복사하여 붙여넣을 수 있는 코드 조각이 포함되어야 합니다. 이슈에 사용자의 로컬 모델 가중치나 로컬 데이터를 사용하는 경우, 독자가 액세스할 수 없는 경우 이슈를 해결할 수 없습니다. 데이터나 모델을 공유할 수 없는 경우, 더미 모델이나 더미 데이터를 만들어 사용해보세요.
4. **간결성**: 가능한 한 간결하게 유지하여 독자가 문제를 빠르게 이해할 수 있도록 도와주세요. 문제와 관련이 없는 코드나 정보는 모두 제거해주세요. 버그를 발견한 경우, 문제를 설명하는 가장 간단한 코드 예제를 만들어보세요. 버그를 발견한 후에는 작업 흐름 전체를 문제에 던지는 것이 아니라, 에러가 발생하는 훈련 코드의 어느 부분이 문제인지 먼저 이해하고 몇 줄로 재현해보세요. 전체 데이터셋 대신 더미 데이터를 사용해보세요.
5. 링크 추가하기. 특정한 이름, 메서드, 또는 모델을 참조하는 경우, 독자가 더 잘 이해할 수 있도록 링크를 제공해주세요. 특정 PR이나 이슈를 참조하는 경우, 해당 이슈에 링크를 걸어주세요. 독자가 무엇을 말하는지 알고 있다고 가정하지 마세요. 이슈에 링크를 추가할수록 좋습니다.
6. 포맷팅. 코드를 파이썬 코드 구문으로, 에러 메시지를 일반 코드 구문으로 형식화하여 이슈를 깔끔하게 작성하세요. 자세한 내용은 [GitHub 공식 포맷팅 문서](https://docs.github.com/en/get-started/writing-on-github/getting-started-with-writing-and-formatting-on-github/basic-writing-and-formatting-syntax)를 참조하세요.
7. 여러분의 이슈를 단순히 해결해야 할 티켓으로 생각하지 말고, 잘 작성된 백과사전 항목으로 생각해보세요. 추가된 모든 이슈는 공개적으로 이용 가능한 지식에 대한 기여입니다. 잘 작성된 이슈를 추가함으로써 메인테이너가 여러분의 이슈를 더 쉽게 해결할 수 있게 할 뿐만 아니라, 전체 커뮤니티가 라이브러리의 특정 측면을 더 잘 이해할 수 있도록 도움을 주게 됩니다.

## 좋은 PR 작성 방법 [[how-to-write-a-good-pr]]

1. 카멜레온이 되세요. 기존의 디자인 패턴과 구문을 이해하고, 여러분이 추가하는 코드가 기존 코드베이스와 자연스럽게 어우러지도록 해야 합니다. 기존 디자인 패턴이나 사용자 인터페이스와 크게 다른 Pull Request들은 병합되지 않습니다.
2. 레이저처럼 집중하세요. Pull Request는 하나의 문제, 오직 하나의 문제만 해결해야 합니다. "이왕 추가하는 김에 다른 문제도 고치자"는 함정에 빠지지 않도록 주의하세요. 여러 개의 관련 없는 문제를 해결하는 한 번에 해결하는 Pull Request들은 검토하기가 훨씬 더 어렵습니다.
3. 도움이 되는 경우, 추가한 내용이 어떻게 사용되는지 예제 코드 조각을 추가해보세요.
4. Pull Request의 제목은 기여 내용을 요약해야 합니다.
5. Pull Request가 이슈를 해결하는 경우, Pull Request의 설명에 이슈 번호를 언급하여 연결되도록 해주세요 (이슈를 참조하는 사람들이 작업 중임을 알 수 있도록).
6. 진행 중인 작업을 나타내려면 제목에 `[WIP]`를 접두사로 붙여주세요. 이는 중복 작업을 피하고, 병합 준비가 된 PR과 구분할 수 있도록 도움이 됩니다.
7. [좋은 이슈를 작성하는 방법](#how-to-write-a-good-issue)에 설명된 대로 텍스트를 구성하고 형식을 지정해보세요.
8. 기존 테스트가 통과하는지 확인하세요
9. 높은 커버리지를 가진 테스트를 추가하세요. 품질 테스트가 없으면 병합할 수 없습니다.
- 새로운 `@slow` 테스트를 추가하는 경우, 다음 명령을 사용하여 통과하는지 확인하세요.
`RUN_SLOW=1 python -m pytest tests/test_my_new_model.py`.
CircleCI는 느린 테스트를 실행하지 않지만, GitHub Actions는 매일 실행합니다!
10. 모든 공개 메서드는 마크다운과 잘 작동하는 정보성 docstring을 가져야 합니다. 예시로 [`pipeline_latent_diffusion.py`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py)를 참조하세요.
11. 리포지토리가 빠르게 성장하고 있기 때문에, 리포지토리에 큰 부담을 주는 파일이 추가되지 않도록 주의해야 합니다. 이미지, 비디오 및 기타 텍스트가 아닌 파일을 포함합니다. 이러한 파일을 배치하기 위해 hf.co 호스팅 `dataset`인 [`hf-internal-testing`](https://huggingface.co/hf-internal-testing) 또는 [huggingface/documentation-images](https://huggingface.co/datasets/huggingface/documentation-images)를 활용하는 것이 우선입니다.
외부 기여인 경우, 이미지를 PR에 추가하고 Hugging Face 구성원에게 이미지를 이 데이터셋으로 이동하도록 요청하세요.

## PR을 열기 위한 방법 [[how-to-open-a-pr]]

코드를 작성하기 전에, 이미 누군가가 같은 작업을 하고 있는지 확인하기 위해 기존의 PR이나 이슈를 검색하는 것이 좋습니다. 확실하지 않은 경우, 피드백을 받기 위해 이슈를 열어보는 것이 항상 좋은 아이디어입니다.

🧨 Diffusers에 기여하기 위해서는 기본적인 `git` 사용법을 알아야 합니다. `git`은 가장 쉬운 도구는 아니지만, 가장 훌륭한 매뉴얼을 가지고 있습니다. 셸에서 `git --help`을 입력하고 즐기세요. 책을 선호하는 경우, [Pro Git](https://git-scm.com/book/en/v2)은 매우 좋은 참고 자료입니다.

다음 단계를 따라 기여를 시작하세요 ([지원되는 Python 버전](https://github.com/huggingface/diffusers/blob/main/setup.py#L244)):

1. 저장소 페이지에서 'Fork' 버튼을 클릭하여 [저장소](https://github.com/huggingface/diffusers)를 포크합니다. 이렇게 하면 코드의 사본이 GitHub 사용자 계정에 생성됩니다.

2. 포크한 저장소를 로컬 디스크에 클론하고, 기본 저장소를 원격으로 추가하세요:

 ```bash
 $ git clone git@github.com:<your GitHub handle>/diffusers.git
 $ cd diffusers
 $ git remote add upstream https://github.com/huggingface/diffusers.git
 ```

3. 개발 변경 사항을 보관할 새로운 브랜치를 생성하세요:

 ```bash
 $ git checkout -b a-descriptive-name-for-my-changes
 ```

`main` 브랜치 위에서 **절대** 작업하지 마세요.

4. 가상 환경에서 다음 명령을 실행하여 개발 환경을 설정하세요:

 ```bash
 $ pip install -e ".[dev]"
 ```

만약 저장소를 이미 클론한 경우, 가장 최신 변경 사항을 가져오기 위해 `git pull`을 실행해야 할 수도 있습니다.

5. 기능을 브랜치에서 개발하세요.

기능을 작업하는 동안 테스트 스위트가 통과되는지 확인해야 합니다. 다음과 같이 변경 사항에 영향을 받는 테스트를 실행해야 합니다:

 ```bash
 $ pytest tests/<TEST_TO_RUN>.py
 ```

테스트를 실행하기 전에 테스트를 위해 필요한 의존성들을 설치하였는지 확인하세요. 다음의 커맨드를 통해서 확인할 수 있습니다:

 ```bash
 $ pip install -e ".[test]"
 ```

다음 명령어로 전체 테스트 묶음 실행할 수도 있지만, Diffusers가 많이 성장하였기 때문에 결과를 적당한 시간 내에 생성하기 위해서는 강력한 컴퓨터가 필요합니다. 다음은 해당 명령어입니다:

 ```bash
 $ make test
 ```

🧨 Diffusers는 소스 코드를 일관되게 포맷팅하기 위해 `black`과 `isort`를 사용합니다. 변경 사항을 적용한 후에는 다음과 같이 자동 스타일 수정 및 코드 검증을 적용할 수 있습니다:


 ```bash
 $ make style
 ```

🧨 Diffusers `ruff`와 몇개의 커스텀 스크립트를 이용하여 코딩 실수를 확인합니다. 품질 제어는 CI에서 작동하지만, 동일한 검사를 다음을 통해서도 할 수 있습니다:

 ```bash
 $ make quality
 ```

변경사항에 대해 만족한다면 `git add`를 사용하여 변경된 파일을 추가하고 `git commit`을 사용하여 변경사항에 대해 로컬상으로 저장한다:

 ```bash
 $ git add modified_file.py
 $ git commit -m "A descriptive message about your changes."
 ```

코드를 정기적으로 원본 저장소와 동기화하는 것은 좋은 아이디어입니다. 이렇게 하면 변경 사항을 빠르게 반영할 수 있습니다:

 ```bash
 $ git pull upstream main
 ```

변경 사항을 계정에 푸시하려면 다음을 사용하세요:

 ```bash
 $ git push -u origin a-descriptive-name-for-my-changes
 ```

6. 만족하셨다면, GitHub에서 포크한 웹페이지로 이동하여 'Pull request'를 클릭하여 변경사항을 프로젝트 메인테이너에게 검토를 요청합니다.

7. 메인테이너가 변경 사항을 요청하는 것은 괜찮습니다. 핵심 기여자들에게도 일어나는 일입니다! 따라서 변경 사항을 Pull request에서 볼 수 있도록 로컬 브랜치에서 작업하고 변경 사항을 포크에 푸시하면 자동으로 Pull request에 나타납니다.

### 테스트 [[tests]]

라이브러리 동작과 여러 예제를 테스트하기 위해 포괄적인 테스트 묶음이 포함되어 있습니다. 라이브러리 테스트는 [tests 폴더](https://github.com/huggingface/diffusers/tree/main/tests)에서 찾을 수 있습니다.

`pytest`와 `pytest-xdist`를 선호하는 이유는 더 빠르기 때문입니다. 루트 디렉토리에서 라이브러리를 위해 `pytest`로 테스트를 실행하는 방법은 다음과 같습니다:

```bash
$ python -m pytest -n auto --dist=loadfile -s -v ./tests/
```

사실, `make test`는 이렇게 구현되어 있습니다!

작업 중인 기능만 테스트하기 위해 더 작은 테스트 세트를 지정할 수 있습니다.

기본적으로 느린 테스트는 건너뜁니다. `RUN_SLOW` 환경 변수를 `yes`로 설정하여 실행할 수 있습니다. 이는 많은 기가바이트의 모델을 다운로드합니다. 충분한 디스크 공간과 좋은 인터넷 연결 또는 많은 인내심이 필요합니다!

```bash
$ RUN_SLOW=yes python -m pytest -n auto --dist=loadfile -s -v ./tests/
```

`unittest`는 완전히 지원됩니다. 다음은 `unittest`를 사용하여 테스트를 실행하는 방법입니다:

```bash
$ python -m unittest discover -s tests -t . -v
$ python -m unittest discover -s examples -t examples -v
```

### upstream(HuggingFace) main과 forked main 동기화하기 [[syncing-forked-main-with-upstream-huggingface-main]]

upstream 저장소에 불필요한 참조 노트를 추가하고 관련 개발자에게 알림을 보내는 것을 피하기 위해,
forked 저장소의 main 브랜치를 동기화할 때 다음 단계를 따르세요:
1. 가능한 경우, forked 저장소에서 브랜치와 PR을 사용하여 upstream과 동기화하는 것을 피하세요. 대신 forked main으로 직접 병합하세요.
2. PR이 절대적으로 필요한 경우, 브랜치를 체크아웃한 후 다음 단계를 사용하세요:
```bash
$ git checkout -b your-branch-for-syncing
$ git pull --squash --no-commit upstream main
$ git commit -m '<your message without GitHub references>'
$ git push --set-upstream origin your-branch-for-syncing
```

### 스타일 가이드 [[style-guide]]

Documentation string에 대해서는, 🧨 Diffusers는 [Google 스타일](https://google.github.io/styleguide/pyguide.html)을 따릅니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/conceptual/contribution.md" />

### 🧨 Diffusers의 윤리 지침 [[-diffusers-ethical-guidelines]]
https://huggingface.co/docs/diffusers/main/ko/conceptual/ethical_guidelines.md

# 🧨 Diffusers의 윤리 지침 [[-diffusers-ethical-guidelines]]

## 서문 [[preamble]]

[Diffusers](https://huggingface.co/docs/diffusers/index)는 사전 훈련된 diffusion 모델을 제공하며, 추론과 훈련을 위한 모듈형 툴박스로 활용됩니다.

이 기술의 실제 적용 사례와 사회에 미칠 수 있는 잠재적 부정적 영향을 고려할 때, Diffusers 라이브러리의 개발, 사용자 기여, 사용에 윤리 지침을 제공하는 것이 중요하다고 생각합니다.

이 기술 사용과 관련된 위험은 여전히 검토 중이지만, 예를 들면: 예술가의 저작권 문제, 딥페이크 악용, 부적절한 맥락에서의 성적 콘텐츠 생성, 비동의 사칭, 소수자 집단 억압을 영속화하는 유해한 사회적 편견 등이 있습니다.
우리는 이러한 위험을 지속적으로 추적하고, 커뮤니티의 반응과 소중한 피드백에 따라 아래 지침을 조정할 것입니다.

## 범위 [[scope]]

Diffusers 커뮤니티는 프로젝트 개발에 다음 윤리 지침을 적용하며, 특히 윤리적 문제와 관련된 민감한 주제에 대해 커뮤니티의 기여를 조정하는 데 도움을 줄 것입니다.

## 윤리 지침 [[ethical-guidelines]]

다음 윤리 지침은 일반적으로 적용되지만, 윤리적으로 민감한 문제와 관련된 기술적 선택을 할 때 우선적으로 적용됩니다. 또한, 해당 기술의 최신 동향과 관련된 새로운 위험이 발생함에 따라 이러한 윤리 원칙을 지속적으로 조정할 것을 약속합니다.

- **투명성**: 우리는 PR 관리, 사용자에게 선택의 이유 설명, 기술적 의사결정 과정에서 투명성을 유지할 것을 약속합니다.

- **일관성**: 프로젝트 관리에서 모든 사용자에게 동일한 수준의 관심을 보장하고, 기술적으로 안정적이고 일관된 상태를 유지할 것을 약속합니다.

- **간결성**: Diffusers 라이브러리를 쉽게 사용하고 활용할 수 있도록, 프로젝트의 목표를 간결하고 일관성 있게 유지할 것을 약속합니다.

- **접근성**: Diffusers 프로젝트는 기술적 전문지식이 없어도 기여할 수 있도록 진입장벽을 낮춥니다. 이를 통해 연구 결과물이 커뮤니티에 더 잘 접근될 수 있습니다.

- **재현성**: 우리는 Diffusers 라이브러리를 통해 제공되는 업스트림 코드, 모델, 데이터셋의 재현성에 대해 투명하게 공개하는 것을 목표로 합니다.

- **책임**: 커뮤니티와 팀워크를 통해, 이 기술의 잠재적 위험을 예측하고 완화하는 데 공동 책임을 집니다.

## 구현 사례: 안전 기능과 메커니즘 [[examples-of-implementations-safety-features-and-mechanisms]]

팀은 diffusion 기술과 관련된 잠재적 윤리 및 사회적 위험에 대응하기 위해 기술적·비기술적 도구를 제공하고자 노력하고 있습니다. 또한, 커뮤니티의 참여는 이러한 기능 구현과 인식 제고에 매우 중요합니다.

- [**커뮤니티 탭**](https://huggingface.co/docs/hub/repositories-pull-requests-discussions): 커뮤니티가 프로젝트에 대해 토론하고 더 나은 협업을 할 수 있도록 지원합니다.

- **편향 탐색 및 평가**: Hugging Face 팀은 Stable Diffusion 모델의 편향성을 대화형으로 보여주는 [space](https://huggingface.co/spaces/society-ethics/DiffusionBiasExplorer)를 제공합니다. 우리는 이러한 편향 탐색과 평가를 지원하고 장려합니다.

- **배포에서의 안전 유도**

  - [**안전한 Stable Diffusion**](https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_diffusion/stable_diffusion_safe): 필터링되지 않은 웹 크롤링 데이터셋으로 훈련된 Stable Diffusion과 같은 모델이 부적절하게 변질되는 문제를 완화합니다. 관련 논문: [Safe Latent Diffusion: Mitigating Inappropriate Degeneration in Diffusion Models](https://huggingface.co/papers/2211.05105).

  - [**안전 검사기**](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/safety_checker.py): 생성된 이미지가 임베딩 공간에서 하드코딩된 유해 개념 클래스와 일치할 확률을 확인하고 비교합니다. 유해 개념은 역공학을 방지하기 위해 의도적으로 숨겨져 있습니다.

- **Hub에서의 단계적 배포**: 특히 민감한 상황에서는 일부 리포지토리에 대한 접근을 제한할 수 있습니다. 단계적 배포는 리포지토리 작성자가 사용에 대해 더 많은 통제권을 갖도록 하는 중간 단계입니다.

- **라이선싱**: [OpenRAILs](https://huggingface.co/blog/open_rail)와 같은 새로운 유형의 라이선스를 통해 자유로운 접근을 보장하면서도 보다 책임 있는 사용을 위한 일련의 제한을 둘 수 있습니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/conceptual/ethical_guidelines.md" />

### Diffusion 모델을 학습하기
https://huggingface.co/docs/diffusers/main/ko/tutorials/basic_training.md

# Diffusion 모델을 학습하기

Unconditional 이미지 생성은 학습에 사용된 데이터셋과 유사한 이미지를 생성하는 diffusion 모델에서 인기 있는 어플리케이션입니다. 일반적으로, 가장 좋은 결과는 특정 데이터셋에 사전 훈련된 모델을 파인튜닝하는 것으로 얻을 수 있습니다. 이 [허브](https://huggingface.co/search/full-text?q=unconditional-image-generation&type=model)에서 이러한 많은 체크포인트를 찾을 수 있지만, 만약 마음에 드는 체크포인트를 찾지 못했다면, 언제든지 스스로 학습할 수 있습니다!

이 튜토리얼은 나만의 🦋 나비 🦋를 생성하기 위해 [Smithsonian Butterflies](https://huggingface.co/datasets/huggan/smithsonian_butterflies_subset) 데이터셋의 하위 집합에서 `UNet2DModel` 모델을 학습하는 방법을 가르쳐줄 것입니다.

> [!TIP]
> 💡 이 학습 튜토리얼은 [Training with 🧨 Diffusers](https://colab.research.google.com/github/huggingface/notebooks/blob/main/diffusers/training_example.ipynb) 노트북 기반으로 합니다. Diffusion 모델의 작동 방식 및 자세한 내용은 노트북을 확인하세요!

시작 전에, 🤗 Datasets을 불러오고 전처리하기 위해 데이터셋이 설치되어 있는지 다수 GPU에서 학습을 간소화하기 위해 🤗 Accelerate 가 설치되어 있는지 확인하세요. 그 후 학습 메트릭을 시각화하기 위해 [TensorBoard](https://www.tensorflow.org/tensorboard)를 또한 설치하세요. (또한 학습 추적을 위해 [Weights & Biases](https://docs.wandb.ai/)를 사용할 수 있습니다.)

```bash
!pip install diffusers[training]
```

커뮤니티에 모델을 공유할 것을 권장하며, 이를 위해서 Hugging Face 계정에 로그인을 해야 합니다. (계정이 없다면 [여기](https://hf.co/join)에서 만들 수 있습니다.) 노트북에서 로그인할 수 있으며 메시지가 표시되면 토큰을 입력할 수 있습니다.

```py
>>> from huggingface_hub import notebook_login

>>> notebook_login()
```

또는 터미널로 로그인할 수 있습니다:

```bash
hf auth login
```

모델 체크포인트가 상당히 크기 때문에 [Git-LFS](https://git-lfs.com/)에서 대용량 파일의 버전 관리를 할 수 있습니다.

```bash
!sudo apt -qq install git-lfs
!git config --global credential.helper store
```


## 학습 구성

편의를 위해 학습 파라미터들을 포함한 `TrainingConfig` 클래스를 생성합니다 (자유롭게 조정 가능):

```py
>>> from dataclasses import dataclass


>>> @dataclass
... class TrainingConfig:
...     image_size = 128  # 생성되는 이미지 해상도
...     train_batch_size = 16
...     eval_batch_size = 16  # 평가 동안에 샘플링할 이미지 수
...     num_epochs = 50
...     gradient_accumulation_steps = 1
...     learning_rate = 1e-4
...     lr_warmup_steps = 500
...     save_image_epochs = 10
...     save_model_epochs = 30
...     mixed_precision = "fp16"  # `no`는 float32, 자동 혼합 정밀도를 위한 `fp16`
...     output_dir = "ddpm-butterflies-128"  # 로컬 및 HF Hub에 저장되는 모델명

...     push_to_hub = True  # 저장된 모델을 HF Hub에 업로드할지 여부
...     hub_private_repo = None
...     overwrite_output_dir = True  # 노트북을 다시 실행할 때 이전 모델에 덮어씌울지
...     seed = 0


>>> config = TrainingConfig()
```


## 데이터셋 불러오기

🤗 Datasets 라이브러리와 [Smithsonian Butterflies](https://huggingface.co/datasets/huggan/smithsonian_butterflies_subset) 데이터셋을 쉽게 불러올 수 있습니다.

```py
>>> from datasets import load_dataset

>>> config.dataset_name = "huggan/smithsonian_butterflies_subset"
>>> dataset = load_dataset(config.dataset_name, split="train")
```

💡[HugGan Community Event](https://huggingface.co/huggan) 에서 추가의 데이터셋을 찾거나 로컬의 [`ImageFolder`](https://huggingface.co/docs/datasets/image_dataset#imagefolder)를 만듦으로써 나만의 데이터셋을 사용할 수 있습니다. HugGan Community Event 에 가져온 데이터셋의 경우 리포지토리의 id로 `config.dataset_name` 을 설정하고, 나만의 이미지를 사용하는 경우 `imagefolder` 를 설정합니다.

🤗 Datasets은 `Image` 기능을 사용해 자동으로 이미지 데이터를 디코딩하고 [`PIL.Image`](https://pillow.readthedocs.io/en/stable/reference/Image.html)로 불러옵니다. 이를 시각화 해보면:

```py
>>> import matplotlib.pyplot as plt

>>> fig, axs = plt.subplots(1, 4, figsize=(16, 4))
>>> for i, image in enumerate(dataset[:4]["image"]):
...     axs[i].imshow(image)
...     axs[i].set_axis_off()
>>> fig.show()
```

![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/butterflies_ds.png)

이미지는 모두 다른 사이즈이기 때문에, 우선 전처리가 필요합니다:

-   `Resize` 는 `config.image_size` 에 정의된 이미지 사이즈로 변경합니다.
-   `RandomHorizontalFlip` 은 랜덤적으로 이미지를 미러링하여 데이터셋을 보강합니다.
-   `Normalize` 는 모델이 예상하는 [-1, 1] 범위로 픽셀 값을 재조정 하는데 중요합니다.

```py
>>> from torchvision import transforms

>>> preprocess = transforms.Compose(
...     [
...         transforms.Resize((config.image_size, config.image_size)),
...         transforms.RandomHorizontalFlip(),
...         transforms.ToTensor(),
...         transforms.Normalize([0.5], [0.5]),
...     ]
... )
```

 학습 도중에 `preprocess` 함수를 적용하려면 🤗 Datasets의 `set_transform` 방법이 사용됩니다.

```py
>>> def transform(examples):
...     images = [preprocess(image.convert("RGB")) for image in examples["image"]]
...     return {"images": images}


>>> dataset.set_transform(transform)
```

이미지의 크기가 조정되었는지 확인하기 위해 이미지를 다시 시각화해보세요. 이제 [DataLoader](https://pytorch.org/docs/stable/data#torch.utils.data.DataLoader)에 데이터셋을 포함해 학습할 준비가 되었습니다!

```py
>>> import torch

>>> train_dataloader = torch.utils.data.DataLoader(dataset, batch_size=config.train_batch_size, shuffle=True)
```


## UNet2DModel 생성하기

🧨 Diffusers에 사전학습된 모델들은 모델 클래스에서 원하는 파라미터로 쉽게 생성할 수 있습니다. 예를 들어, `UNet2DModel`를 생성하려면:

```py
>>> from diffusers import UNet2DModel

>>> model = UNet2DModel(
...     sample_size=config.image_size,  # 타겟 이미지 해상도
...     in_channels=3,  # 입력 채널 수, RGB 이미지에서 3
...     out_channels=3,  # 출력 채널 수
...     layers_per_block=2,  # UNet 블럭당 몇 개의 ResNet 레이어가 사용되는지
...     block_out_channels=(128, 128, 256, 256, 512, 512),  # 각 UNet 블럭을 위한 출력 채널 수
...     down_block_types=(
...         "DownBlock2D",  # 일반적인 ResNet 다운샘플링 블럭
...         "DownBlock2D",
...         "DownBlock2D",
...         "DownBlock2D",
...         "AttnDownBlock2D",  # spatial self-attention이 포함된 일반적인 ResNet 다운샘플링 블럭
...         "DownBlock2D",
...     ),
...     up_block_types=(
...         "UpBlock2D",  # 일반적인 ResNet 업샘플링 블럭
...         "AttnUpBlock2D",  # spatial self-attention이 포함된 일반적인 ResNet 업샘플링 블럭
...         "UpBlock2D",
...         "UpBlock2D",
...         "UpBlock2D",
...         "UpBlock2D",
...     ),
... )
```

샘플의 이미지 크기와 모델 출력 크기가 맞는지 빠르게 확인하기 위한 좋은 아이디어가 있습니다:

```py
>>> sample_image = dataset[0]["images"].unsqueeze(0)
>>> print("Input shape:", sample_image.shape)
Input shape: torch.Size([1, 3, 128, 128])

>>> print("Output shape:", model(sample_image, timestep=0).sample.shape)
Output shape: torch.Size([1, 3, 128, 128])
```

훌륭해요! 다음, 이미지에 약간의 노이즈를 더하기 위해 스케줄러가 필요합니다.


## 스케줄러 생성하기

스케줄러는 모델을 학습 또는 추론에 사용하는지에 따라 다르게 작동합니다. 추론시에, 스케줄러는 노이즈로부터 이미지를 생성합니다. 학습시 스케줄러는 diffusion 과정에서의 특정 포인트로부터 모델의 출력 또는 샘플을 가져와 *노이즈 스케줄* 과 *업데이트 규칙*에 따라 이미지에 노이즈를 적용합니다.

`DDPMScheduler`를 보면 이전으로부터 `sample_image`에 랜덤한 노이즈를 더하는 `add_noise` 메서드를 사용합니다:

```py
>>> import torch
>>> from PIL import Image
>>> from diffusers import DDPMScheduler

>>> noise_scheduler = DDPMScheduler(num_train_timesteps=1000)
>>> noise = torch.randn(sample_image.shape)
>>> timesteps = torch.LongTensor([50])
>>> noisy_image = noise_scheduler.add_noise(sample_image, noise, timesteps)

>>> Image.fromarray(((noisy_image.permute(0, 2, 3, 1) + 1.0) * 127.5).type(torch.uint8).numpy()[0])
```

![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/noisy_butterfly.png)

모델의 학습 목적은 이미지에 더해진 노이즈를 예측하는 것입니다. 이 단계에서 손실은 다음과 같이 계산될 수 있습니다:

```py
>>> import torch.nn.functional as F

>>> noise_pred = model(noisy_image, timesteps).sample
>>> loss = F.mse_loss(noise_pred, noise)
```

## 모델 학습하기

지금까지, 모델 학습을 시작하기 위해 많은 부분을 갖추었으며 이제 남은 것은 모든 것을 조합하는 것입니다.

우선 옵티마이저(optimizer)와 학습률 스케줄러(learning rate scheduler)가 필요할 것입니다:

```py
>>> from diffusers.optimization import get_cosine_schedule_with_warmup

>>> optimizer = torch.optim.AdamW(model.parameters(), lr=config.learning_rate)
>>> lr_scheduler = get_cosine_schedule_with_warmup(
...     optimizer=optimizer,
...     num_warmup_steps=config.lr_warmup_steps,
...     num_training_steps=(len(train_dataloader) * config.num_epochs),
... )
```

그 후, 모델을 평가하는 방법이 필요합니다. 평가를 위해, `DDPMPipeline`을 사용해 배치의 이미지 샘플들을 생성하고 그리드 형태로 저장할 수 있습니다:

```py
>>> from diffusers import DDPMPipeline
>>> import math
>>> import os


>>> def make_grid(images, rows, cols):
...     w, h = images[0].size
...     grid = Image.new("RGB", size=(cols * w, rows * h))
...     for i, image in enumerate(images):
...         grid.paste(image, box=(i % cols * w, i // cols * h))
...     return grid


>>> def evaluate(config, epoch, pipeline):
...     # 랜덤한 노이즈로 부터 이미지를 추출합니다.(이는 역전파 diffusion 과정입니다.)
...     # 기본 파이프라인 출력 형태는 `List[PIL.Image]` 입니다.
...     images = pipeline(
...         batch_size=config.eval_batch_size,
...         generator=torch.manual_seed(config.seed),
...     ).images

...     # 이미지들을 그리드로 만들어줍니다.
...     image_grid = make_grid(images, rows=4, cols=4)

...     # 이미지들을 저장합니다.
...     test_dir = os.path.join(config.output_dir, "samples")
...     os.makedirs(test_dir, exist_ok=True)
...     image_grid.save(f"{test_dir}/{epoch:04d}.png")
```

TensorBoard에 로깅, 그래디언트 누적 및 혼합 정밀도 학습을 쉽게 수행하기 위해 🤗 Accelerate를 학습 루프에 함께 앞서 말한 모든 구성 정보들을 묶어 진행할 수 있습니다. 허브에 모델을 업로드 하기 위해 리포지토리 이름 및 정보를 가져오기 위한 함수를 작성하고 허브에 업로드할 수 있습니다.

💡아래의 학습 루프는 어렵고 길어 보일 수 있지만, 나중에 한 줄의 코드로 학습을 한다면 그만한 가치가 있을 것입니다! 만약 기다리지 못하고 이미지를 생성하고 싶다면, 아래 코드를 자유롭게 붙여넣고 작동시키면 됩니다. 🤗

```py
>>> from accelerate import Accelerator
>>> from huggingface_hub import create_repo, upload_folder
>>> from tqdm.auto import tqdm
>>> from pathlib import Path
>>> import os


>>> def train_loop(config, model, noise_scheduler, optimizer, train_dataloader, lr_scheduler):
...     # Initialize accelerator and tensorboard logging
...     accelerator = Accelerator(
...         mixed_precision=config.mixed_precision,
...         gradient_accumulation_steps=config.gradient_accumulation_steps,
...         log_with="tensorboard",
...         project_dir=os.path.join(config.output_dir, "logs"),
...     )
...     if accelerator.is_main_process:
...         if config.output_dir is not None:
...             os.makedirs(config.output_dir, exist_ok=True)
...         if config.push_to_hub:
...             repo_id = create_repo(
...                 repo_id=config.hub_model_id or Path(config.output_dir).name, exist_ok=True
...             ).repo_id
...         accelerator.init_trackers("train_example")

...     # 모든 것이 준비되었습니다.
...     # 기억해야 할 특정한 순서는 없으며 준비한 방법에 제공한 것과 동일한 순서로 객체의 압축을 풀면 됩니다.
...     model, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
...         model, optimizer, train_dataloader, lr_scheduler
...     )

...     global_step = 0

...     # 이제 모델을 학습합니다.
...     for epoch in range(config.num_epochs):
...         progress_bar = tqdm(total=len(train_dataloader), disable=not accelerator.is_local_main_process)
...         progress_bar.set_description(f"Epoch {epoch}")

...         for step, batch in enumerate(train_dataloader):
...             clean_images = batch["images"]
...             # 이미지에 더할 노이즈를 샘플링합니다.
...             noise = torch.randn(clean_images.shape, device=clean_images.device)
...             bs = clean_images.shape[0]

...             # 각 이미지를 위한 랜덤한 타임스텝(timestep)을 샘플링합니다.
...             timesteps = torch.randint(
...                 0, noise_scheduler.config.num_train_timesteps, (bs,), device=clean_images.device,
...                 dtype=torch.int64
...             )

...             # 각 타임스텝의 노이즈 크기에 따라 깨끗한 이미지에 노이즈를 추가합니다.
...             # (이는 foward diffusion 과정입니다.)
...             noisy_images = noise_scheduler.add_noise(clean_images, noise, timesteps)

...             with accelerator.accumulate(model):
...                 # 노이즈를 반복적으로 예측합니다.
...                 noise_pred = model(noisy_images, timesteps, return_dict=False)[0]
...                 loss = F.mse_loss(noise_pred, noise)
...                 accelerator.backward(loss)

...                 accelerator.clip_grad_norm_(model.parameters(), 1.0)
...                 optimizer.step()
...                 lr_scheduler.step()
...                 optimizer.zero_grad()

...             progress_bar.update(1)
...             logs = {"loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0], "step": global_step}
...             progress_bar.set_postfix(**logs)
...             accelerator.log(logs, step=global_step)
...             global_step += 1

...         # 각 에포크가 끝난 후 evaluate()와 몇 가지 데모 이미지를 선택적으로 샘플링하고 모델을 저장합니다.
...         if accelerator.is_main_process:
...             pipeline = DDPMPipeline(unet=accelerator.unwrap_model(model), scheduler=noise_scheduler)

...             if (epoch + 1) % config.save_image_epochs == 0 or epoch == config.num_epochs - 1:
...                 evaluate(config, epoch, pipeline)

...             if (epoch + 1) % config.save_model_epochs == 0 or epoch == config.num_epochs - 1:
...                 if config.push_to_hub:
...                     upload_folder(
...                         repo_id=repo_id,
...                         folder_path=config.output_dir,
...                         commit_message=f"Epoch {epoch}",
...                         ignore_patterns=["step_*", "epoch_*"],
...                     )
...                 else:
...                     pipeline.save_pretrained(config.output_dir)
```

휴, 코드가 꽤 많았네요! 하지만 🤗 Accelerate의 `notebook_launcher` 함수와 학습을 시작할 준비가 되었습니다. 함수에 학습 루프, 모든 학습 인수, 학습에 사용할 프로세스 수(사용 가능한 GPU의 수를 변경할 수 있음)를 전달합니다:

```py
>>> from accelerate import notebook_launcher

>>> args = (config, model, noise_scheduler, optimizer, train_dataloader, lr_scheduler)

>>> notebook_launcher(train_loop, args, num_processes=1)
```

한번 학습이 완료되면, diffusion 모델로 생성된 최종 🦋이미지🦋를 확인해보길 바랍니다!

```py
>>> import glob

>>> sample_images = sorted(glob.glob(f"{config.output_dir}/samples/*.png"))
>>> Image.open(sample_images[-1])
```

![](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/butterflies_final.png)

## 다음 단계

Unconditional 이미지 생성은 학습될 수 있는 작업 중 하나의 예시입니다. 다른 작업과 학습 방법은 [🧨 Diffusers 학습 예시](../training/overview) 페이지에서 확인할 수 있습니다. 다음은 학습할 수 있는 몇 가지 예시입니다:

-   [Textual Inversion](../training/text_inversion), 특정 시각적 개념을 학습시켜 생성된 이미지에 통합시키는 알고리즘입니다.
-   [DreamBooth](../training/dreambooth), 주제에 대한 몇 가지 입력 이미지들이 주어지면 주제에 대한 개인화된 이미지를 생성하기 위한 기술입니다.
-   [Guide](../training/text2image) 데이터셋에 Stable Diffusion 모델을 파인튜닝하는 방법입니다.
-   [Guide](../training/lora)  LoRA를 사용해 매우 큰 모델을 빠르게 파인튜닝하기 위한 메모리 효율적인 기술입니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/tutorials/basic_training.md" />

### Overview
https://huggingface.co/docs/diffusers/main/ko/tutorials/tutorial_overview.md

# Overview

🧨 Diffusers에 오신 걸 환영합니다! 여러분이 diffusion 모델과 생성 AI를 처음 접하고, 더 많은 걸 배우고 싶으셨다면 제대로 찾아오셨습니다. 이 튜토리얼은 diffusion model을 여러분에게 젠틀하게 소개하고, 라이브러리의 기본 사항(핵심 구성요소와 🧨 Diffusers 사용법)을 이해하는 데 도움이 되도록 설계되었습니다.

여러분은 이 튜토리얼을 통해 빠르게 생성하기 위해선 추론 파이프라인을 어떻게 사용해야 하는지, 그리고 라이브러리를 modular toolbox처럼 이용해서 여러분만의 diffusion system을 구축할 수 있도록 파이프라인을 분해하는 법을 배울 수 있습니다. 다음 단원에서는 여러분이 원하는 것을 생성하기 위해 자신만의 diffusion model을 학습하는 방법을 배우게 됩니다.

튜토리얼을 완료한다면 여러분은 라이브러리를 직접 탐색하고, 자신의 프로젝트와 애플리케이션에 적용할 스킬들을 습득할 수 있을 겁니다.

[Discord](https://discord.com/invite/JfAtkvEtRb)나 [포럼](https://discuss.huggingface.co/c/discussion-related-to-httpsgithubcomhuggingfacediffusers/63) 커뮤니티에 자유롭게 참여해서 다른 사용자와 개발자들과 교류하고 협업해 보세요!

자 지금부터 diffusing을 시작해 보겠습니다! 🧨

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/tutorials/tutorial_overview.md" />

### Intel Gaudi에서 Stable Diffusion을 사용하는 방법
https://huggingface.co/docs/diffusers/main/ko/optimization/habana.md

# Intel Gaudi에서 Stable Diffusion을 사용하는 방법

🤗 Diffusers는 🤗 [Optimum Habana](https://huggingface.co/docs/optimum/habana/usage_guides/stable_diffusion)를 통해서 Habana Gaudi와 호환됩니다.

## 요구 사항

- Optimum Habana 1.4 또는 이후, [여기](https://huggingface.co/docs/optimum/habana/installation)에 설치하는 방법이 있습니다.
- SynapseAI 1.8.


## 추론 파이프라인

Gaudi에서 Stable Diffusion 1 및 2로 이미지를 생성하려면 두 인스턴스를 인스턴스화해야 합니다:
- [`GaudiStableDiffusionPipeline`](https://huggingface.co/docs/optimum/habana/package_reference/stable_diffusion_pipeline)이 포함된 파이프라인. 이 파이프라인은 *텍스트-이미지 생성*을 지원합니다.
- [`GaudiDDIMScheduler`](https://huggingface.co/docs/optimum/habana/package_reference/stable_diffusion_pipeline#optimum.habana.diffusers.GaudiDDIMScheduler)이 포함된 스케줄러. 이 스케줄러는 Habana Gaudi에 최적화되어 있습니다.

파이프라인을 초기화할 때, HPU에 배포하기 위해 `use_habana=True`를 지정해야 합니다.
또한 가능한 가장 빠른 생성을 위해 `use_hpu_graphs=True`로 **HPU 그래프**를 활성화해야 합니다.
마지막으로, [Hugging Face Hub](https://huggingface.co/Habana)에서 다운로드할 수 있는 [Gaudi configuration](https://huggingface.co/docs/optimum/habana/package_reference/gaudi_config)을 지정해야 합니다.

```python
from optimum.habana import GaudiConfig
from optimum.habana.diffusers import GaudiDDIMScheduler, GaudiStableDiffusionPipeline

model_name = "stabilityai/stable-diffusion-2-base"
scheduler = GaudiDDIMScheduler.from_pretrained(model_name, subfolder="scheduler")
pipeline = GaudiStableDiffusionPipeline.from_pretrained(
    model_name,
    scheduler=scheduler,
    use_habana=True,
    use_hpu_graphs=True,
    gaudi_config="Habana/stable-diffusion",
)
```

파이프라인을 호출하여 하나 이상의 프롬프트에서 배치별로 이미지를 생성할 수 있습니다.

```python
outputs = pipeline(
    prompt=[
        "High quality photo of an astronaut riding a horse in space",
        "Face of a yellow cat, high resolution, sitting on a park bench",
    ],
    num_images_per_prompt=10,
    batch_size=4,
)
```

더 많은 정보를 얻기 위해, Optimum Habana의 [문서](https://huggingface.co/docs/optimum/habana/usage_guides/stable_diffusion)와 공식 GitHub 저장소에 제공된 [예시](https://github.com/huggingface/optimum-habana/tree/main/examples/stable-diffusion)를 확인하세요.


## 벤치마크

다음은 [Habana/stable-diffusion](https://huggingface.co/Habana/stable-diffusion) Gaudi 구성(혼합 정밀도 bf16/fp32)을 사용하는 Habana first-generation Gaudi 및 Gaudi2의 지연 시간입니다:

|                        | Latency (배치 크기 = 1) | Throughput (배치 크기 = 8) |
| ---------------------- |:------------------------:|:---------------------------:|
| first-generation Gaudi | 4.29s                    | 0.283 images/s              |
| Gaudi2                 | 1.54s                    | 0.904 images/s              |


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/habana.md" />

### Token Merging (토큰 병합)
https://huggingface.co/docs/diffusers/main/ko/optimization/tome.md

# Token Merging (토큰 병합)

Token Merging (introduced in [Token Merging: Your ViT But Faster](https://huggingface.co/papers/2210.09461))은 트랜스포머 기반 네트워크의 forward pass에서 중복 토큰이나 패치를 점진적으로 병합하는 방식으로 작동합니다. 이를 통해 기반 네트워크의 추론 지연 시간을 단축할 수 있습니다.

Token Merging(ToMe)이 출시된 후, 저자들은 [Fast Stable Diffusion을 위한 토큰 병합](https://huggingface.co/papers/2303.17604)을 발표하여 Stable Diffusion과 더 잘 호환되는 ToMe 버전을 소개했습니다. ToMe를 사용하면 `DiffusionPipeline`의 추론 지연 시간을 부드럽게 단축할 수 있습니다. 이 문서에서는 ToMe를 `StableDiffusionPipeline`에 적용하는 방법, 예상되는 속도 향상, `StableDiffusionPipeline`에서 ToMe를 사용할 때의 질적 측면에 대해 설명합니다.

## ToMe 사용하기

ToMe의 저자들은 [`tomesd`](https://github.com/dbolya/tomesd)라는 편리한 Python 라이브러리를 공개했는데, 이 라이브러리를 이용하면 `DiffusionPipeline`에 ToMe를 다음과 같이 적용할 수 있습니다:

```diff
from diffusers import StableDiffusionPipeline
import tomesd

pipeline = StableDiffusionPipeline.from_pretrained(
      "stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16
).to("cuda")
+ tomesd.apply_patch(pipeline, ratio=0.5)

image = pipeline("a photo of an astronaut riding a horse on mars").images[0]
```

이것이 다입니다!

`tomesd.apply_patch()`는 파이프라인 추론 속도와 생성된 토큰의 품질 사이의 균형을 맞출 수 있도록 [여러 개의 인자](https://github.com/dbolya/tomesd#usage)를 노출합니다. 이러한 인수 중 가장 중요한 것은 `ratio(비율)`입니다. `ratio`은 forward pass 중에 병합될 토큰의 수를 제어합니다. `tomesd`에 대한 자세한 내용은 해당 리포지토리(https://github.com/dbolya/tomesd) 및 [논문](https://huggingface.co/papers/2303.17604)을 참고하시기 바랍니다.

## `StableDiffusionPipeline`으로 `tomesd` 벤치마킹하기

We benchmarked the impact of using `tomesd` on `StableDiffusionPipeline` along with [xformers](https://huggingface.co/docs/diffusers/optimization/xformers) across different image resolutions. We used A100 and V100 as our test GPU devices with the following development environment (with Python 3.8.5):
다양한 이미지 해상도에서 [xformers](https://huggingface.co/docs/diffusers/optimization/xformers)를 적용한 상태에서, `StableDiffusionPipeline`에 `tomesd`를 사용했을 때의 영향을 벤치마킹했습니다. 테스트 GPU 장치로 A100과 V100을 사용했으며 개발 환경은 다음과 같습니다(Python 3.8.5 사용):

```bash
- `diffusers` version: 0.15.1
- Python version: 3.8.16
- PyTorch version (GPU?): 1.13.1+cu116 (True)
- Huggingface_hub version: 0.13.2
- Transformers version: 4.27.2
- Accelerate version: 0.18.0
- xFormers version: 0.0.16
- tomesd version: 0.1.2
```

벤치마킹에는 다음 스크립트를 사용했습니다: [https://gist.github.com/sayakpaul/27aec6bca7eb7b0e0aa4112205850335](https://gist.github.com/sayakpaul/27aec6bca7eb7b0e0aa4112205850335). 결과는 다음과 같습니다:

### A100

| 해상도 | 배치 크기 | Vanilla | ToMe | ToMe + xFormers | ToMe 속도 향상 (%) | ToMe + xFormers 속도 향상 (%) |
| --- | --- | --- | --- | --- | --- | --- |
| 512 | 10 | 6.88 | 5.26 | 4.69 | 23.54651163 | 31.83139535 |
|  |  |  |  |  |  |  |
| 768 | 10 | OOM | 14.71 | 11 |  |  |
|  | 8 | OOM | 11.56 | 8.84 |  |  |
|  | 4 | OOM | 5.98 | 4.66 |  |  |
|  | 2 | 4.99 | 3.24 | 3.1 | 35.07014028 | 37.8757515 |
|  | 1 | 3.29 | 2.24 | 2.03 | 31.91489362 | 38.29787234 |
|  |  |  |  |  |  |  |
| 1024 | 10 | OOM | OOM | OOM |  |  |
|  | 8 | OOM | OOM | OOM |  |  |
|  | 4 | OOM | 12.51 | 9.09 |  |  |
|  | 2 | OOM | 6.52 | 4.96 |  |  |
|  | 1 | 6.4 | 3.61 | 2.81 | 43.59375 | 56.09375 |

***결과는 초 단위입니다. 속도 향상은 `Vanilla`과 비교해 계산됩니다.***

### V100

| 해상도 | 배치 크기 | Vanilla | ToMe | ToMe + xFormers | ToMe 속도 향상 (%) | ToMe + xFormers 속도 향상 (%) |
| --- | --- | --- | --- | --- | --- | --- |
| 512 | 10 | OOM | 10.03 | 9.29 |  |  |
|  | 8 | OOM | 8.05 | 7.47 |  |  |
|  | 4 | 5.7 | 4.3 | 3.98 | 24.56140351 | 30.1754386 |
|  | 2 | 3.14 | 2.43 | 2.27 | 22.61146497 | 27.70700637 |
|  | 1 | 1.88 | 1.57 | 1.57 | 16.4893617 | 16.4893617 |
|  |  |  |  |  |  |  |
| 768 | 10 | OOM | OOM | 23.67 |  |  |
|  | 8 | OOM | OOM | 18.81 |  |  |
|  | 4 | OOM | 11.81 | 9.7 |  |  |
|  | 2 | OOM | 6.27 | 5.2 |  |  |
|  | 1 | 5.43 | 3.38 | 2.82 | 37.75322284 | 48.06629834 |
|  |  |  |  |  |  |  |
| 1024 | 10 | OOM | OOM | OOM |  |  |
|  | 8 | OOM | OOM | OOM |  |  |
|  | 4 | OOM | OOM | 19.35 |  |  |
|  | 2 | OOM | 13 | 10.78 |  |  |
|  | 1 | OOM | 6.66 | 5.54 |  |  |

위의 표에서 볼 수 있듯이, 이미지 해상도가 높을수록 `tomesd`를 사용한 속도 향상이 더욱 두드러집니다. 또한 `tomesd`를 사용하면 1024x1024와 같은 더 높은 해상도에서 파이프라인을 실행할 수 있다는 점도 흥미롭습니다.

[`torch.compile()`](https://huggingface.co/docs/diffusers/optimization/torch2.0)을 사용하면 추론 속도를 더욱 높일 수 있습니다.

## 품질

As reported in [the paper](https://huggingface.co/papers/2303.17604), ToMe can preserve the quality of the generated images to a great extent while speeding up inference. By increasing the `ratio`, it is possible to further speed up inference, but that might come at the cost of a deterioration in the image quality.

To test the quality of the generated samples using our setup, we sampled a few prompts from the “Parti Prompts” (introduced in [Parti](https://parti.research.google/)) and performed inference with the `StableDiffusionPipeline` in the following settings:

[논문](https://huggingface.co/papers/2303.17604)에 보고된 바와 같이, ToMe는 생성된 이미지의 품질을 상당 부분 보존하면서 추론 속도를 높일 수 있습니다. `ratio`을 높이면 추론 속도를 더 높일 수 있지만, 이미지 품질이 저하될 수 있습니다.

해당 설정을 사용하여 생성된 샘플의 품질을 테스트하기 위해, "Parti 프롬프트"([Parti](https://parti.research.google/)에서 소개)에서 몇 가지 프롬프트를 샘플링하고 다음 설정에서 `StableDiffusionPipeline`을 사용하여 추론을 수행했습니다:

- Vanilla `StableDiffusionPipeline`
- `StableDiffusionPipeline` + ToMe
- `StableDiffusionPipeline` + ToMe + xformers

생성된 샘플의 품질이 크게 저하되는 것을 발견하지 못했습니다. 다음은 샘플입니다:

![tome-samples](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/tome/tome_samples.png)

생성된 샘플은 [여기](https://wandb.ai/sayakpaul/tomesd-results/runs/23j4bj3i?workspace=)에서 확인할 수 있습니다. 이 실험을 수행하기 위해 [이 스크립트](https://gist.github.com/sayakpaul/8cac98d7f22399085a060992f411ecbd)를 사용했습니다.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/tome.md" />

### xFormers 설치하기
https://huggingface.co/docs/diffusers/main/ko/optimization/xformers.md

# xFormers 설치하기

추론과 학습 모두에 [xFormers](https://github.com/facebookresearch/xformers)를 사용하는 것이 좋습니다.
자체 테스트로 어텐션 블록에서 수행된 최적화가 더 빠른 속도와 적은 메모리 소비를 확인했습니다.

2023년 1월에 출시된 xFormers 버전 '0.0.16'부터 사전 빌드된 pip wheel을 사용하여 쉽게 설치할 수 있습니다:

```bash
pip install xformers
```

> [!TIP]
> xFormers PIP 패키지에는 최신 버전의 PyTorch(xFormers 0.0.16에 1.13.1)가 필요합니다. 이전 버전의 PyTorch를 사용해야 하는 경우 [프로젝트 지침](https://github.com/facebookresearch/xformers#installing-xformers)의 소스를 사용해 xFormers를 설치하는 것이 좋습니다.

xFormers를 설치하면, [여기](fp16#memory-efficient-attention)서 설명한 것처럼 'enable_xformers_memory_efficient_attention()'을 사용하여 추론 속도를 높이고 메모리 소비를 줄일 수 있습니다.

> [!WARNING]
> [이 이슈](https://github.com/huggingface/diffusers/issues/2234#issuecomment-1416931212)에 따르면 xFormers `v0.0.16`에서 GPU를 사용한 학습(파인 튜닝 또는 Dreambooth)을 할 수 없습니다. 해당 문제가 발견되면. 해당 코멘트를 참고해 development 버전을 설치하세요.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/xformers.md" />

### 메모리와 속도
https://huggingface.co/docs/diffusers/main/ko/optimization/fp16.md

# 메모리와 속도

메모리 또는 속도에 대해 🤗 Diffusers *추론*을 최적화하기 위한 몇 가지 기술과 아이디어를 제시합니다.
일반적으로, memory-efficient attention을 위해 [xFormers](https://github.com/facebookresearch/xformers) 사용을 추천하기 때문에, 추천하는 [설치 방법](xformers)을 보고 설치해 보세요.

다음 설정이 성능과 메모리에 미치는 영향에 대해 설명합니다.

|                  | 지연시간  | 속도 향상 |
| ---------------- | ------- | ------- |
| 별도 설정 없음      | 9.50s   | x1      |
| cuDNN auto-tuner | 9.37s   | x1.01   |
| fp16             | 3.61s   | x2.63   |
| Channels Last 메모리 형식     | 3.30s   | x2.88   |
| traced UNet      | 3.21s   | x2.96   |
| memory-efficient attention | 2.63s  | x3.61   |

<em>
   NVIDIA TITAN RTX에서 50 DDIM 스텝의 "a photo of an astronaut riding a horse on mars" 프롬프트로 512x512 크기의 단일 이미지를 생성하였습니다.
</em>

## cuDNN auto-tuner 활성화하기

[NVIDIA cuDNN](https://developer.nvidia.com/cudnn)은 컨볼루션을 계산하는 많은 알고리즘을 지원합니다. Autotuner는 짧은 벤치마크를 실행하고 주어진 입력 크기에 대해 주어진 하드웨어에서 최고의 성능을 가진 커널을 선택합니다.

**컨볼루션 네트워크**를 활용하고 있기 때문에 (다른 유형들은 현재 지원되지 않음), 다음 설정을 통해 추론 전에 cuDNN autotuner를 활성화할 수 있습니다:

```python
import torch

torch.backends.cudnn.benchmark = True
```

### fp32 대신 tf32 사용하기  (Ampere 및 이후 CUDA 장치들에서)

Ampere 및 이후 CUDA 장치에서 행렬곱 및 컨볼루션은 TensorFloat32(TF32) 모드를 사용하여 더 빠르지만 약간 덜 정확할 수 있습니다.
기본적으로 PyTorch는 컨볼루션에 대해 TF32 모드를 활성화하지만 행렬 곱셈은 활성화하지 않습니다.
네트워크에 완전한 float32 정밀도가 필요한 경우가 아니면 행렬 곱셈에 대해서도 이 설정을 활성화하는 것이 좋습니다.
이는 일반적으로 무시할 수 있는 수치의 정확도 손실이 있지만, 계산 속도를 크게 높일 수 있습니다.
그것에 대해 [여기](https://huggingface.co/docs/transformers/v4.18.0/en/performance#tf32)서 더 읽을 수 있습니다.
추론하기 전에 다음을 추가하기만 하면 됩니다:

```python
import torch

torch.backends.cuda.matmul.allow_tf32 = True
```

## 반정밀도 가중치

더 많은 GPU 메모리를 절약하고 더 빠른 속도를 얻기 위해 모델 가중치를 반정밀도(half precision)로 직접 불러오고 실행할 수 있습니다.
여기에는 `fp16`이라는 브랜치에 저장된 float16 버전의 가중치를 불러오고, 그 때 `float16` 유형을 사용하도록 PyTorch에 지시하는 작업이 포함됩니다.

```Python
pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",

    torch_dtype=torch.float16,
)
pipe = pipe.to("cuda")

prompt = "a photo of an astronaut riding a horse on mars"
image = pipe(prompt).images[0]
```

> [!WARNING]
> 어떤 파이프라인에서도 [`torch.autocast`](https://pytorch.org/docs/stable/amp.html#torch.autocast) 를 사용하는 것은 검은색 이미지를 생성할 수 있고, 순수한 float16 정밀도를 사용하는 것보다 항상 느리기 때문에 사용하지 않는 것이 좋습니다.

## 추가 메모리 절약을 위한 슬라이스 어텐션

추가 메모리 절약을 위해, 한 번에 모두 계산하는 대신 단계적으로 계산을 수행하는 슬라이스 버전의 어텐션(attention)을 사용할 수 있습니다.

> [!TIP]
> Attention slicing은 모델이 하나 이상의 어텐션 헤드를 사용하는 한, 배치 크기가 1인 경우에도 유용합니다.
>   하나 이상의 어텐션 헤드가 있는 경우 *QK^T* 어텐션 매트릭스는 상당한 양의 메모리를 절약할 수 있는 각 헤드에 대해 순차적으로 계산될 수 있습니다.

각 헤드에 대해 순차적으로 어텐션 계산을 수행하려면, 다음과 같이 추론 전에 파이프라인에서 `enable_attention_slicing()`를 호출하면 됩니다:

```Python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",

    torch_dtype=torch.float16,
)
pipe = pipe.to("cuda")

prompt = "a photo of an astronaut riding a horse on mars"
pipe.enable_attention_slicing()
image = pipe(prompt).images[0]
```

추론 시간이 약 10% 느려지는 약간의 성능 저하가 있지만 이 방법을 사용하면 3.2GB 정도의 작은 VRAM으로도 Stable Diffusion을 사용할 수 있습니다!


## 더 큰 배치를 위한 sliced VAE 디코드

제한된 VRAM에서 대규모 이미지 배치를 디코딩하거나 32개 이상의 이미지가 포함된 배치를 활성화하기 위해, 배치의 latent 이미지를 한 번에 하나씩 디코딩하는 슬라이스 VAE 디코드를 사용할 수 있습니다.

이를 `enable_attention_slicing()` 또는 `enable_xformers_memory_efficient_attention()`과 결합하여 메모리 사용을 추가로 최소화할 수 있습니다.

VAE 디코드를 한 번에 하나씩 수행하려면 추론 전에 파이프라인에서 `enable_vae_slicing()`을 호출합니다. 예를 들어:

```Python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",

    torch_dtype=torch.float16,
)
pipe = pipe.to("cuda")

prompt = "a photo of an astronaut riding a horse on mars"
pipe.enable_vae_slicing()
images = pipe([prompt] * 32).images
```

다중 이미지 배치에서 VAE 디코드가 약간의 성능 향상이 이루어집니다. 단일 이미지 배치에서는 성능 영향은 없습니다.


<a name="sequential_offloading"></a>
## 메모리 절약을 위해 가속 기능을 사용하여 CPU로 오프로딩

추가 메모리 절약을 위해 가중치를 CPU로 오프로드하고 순방향 전달을 수행할 때만 GPU로 로드할 수 있습니다.

CPU 오프로딩을 수행하려면 `enable_sequential_cpu_offload()`를 호출하기만 하면 됩니다:

```Python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",

    torch_dtype=torch.float16,
)

prompt = "a photo of an astronaut riding a horse on mars"
pipe.enable_sequential_cpu_offload()
image = pipe(prompt).images[0]
```

그러면 메모리 소비를 3GB 미만으로 줄일 수 있습니다.

참고로 이 방법은 전체 모델이 아닌 서브모듈 수준에서 작동합니다. 이는 메모리 소비를 최소화하는 가장 좋은 방법이지만 프로세스의 반복적 특성으로 인해 추론 속도가 훨씬 느립니다. 파이프라인의 UNet 구성 요소는 여러 번 실행됩니다('num_inference_steps' 만큼). 매번 UNet의 서로 다른 서브모듈이 순차적으로 온로드된 다음 필요에 따라 오프로드되므로 메모리 이동 횟수가 많습니다.

> [!TIP]
> 또 다른 최적화 방법인 <a href="#model_offloading">모델 오프로딩</a>을 사용하는 것을 고려하십시오. 이는 훨씬 빠르지만 메모리 절약이 크지는 않습니다.

또한 ttention slicing과 연결해서 최소 메모리(< 2GB)로도 동작할 수 있습니다.


```Python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",

    torch_dtype=torch.float16,
)

prompt = "a photo of an astronaut riding a horse on mars"
pipe.enable_sequential_cpu_offload()
pipe.enable_attention_slicing(1)

image = pipe(prompt).images[0]
```

**참고**: 'enable_sequential_cpu_offload()'를 사용할 때, 미리 파이프라인을 CUDA로 이동하지 **않는** 것이 중요합니다.그렇지 않으면 메모리 소비의 이득이 최소화됩니다. 더 많은 정보를 위해 [이 이슈](https://github.com/huggingface/diffusers/issues/1934)를 보세요.

<a name="model_offloading"></a>
## 빠른 추론과 메모리 메모리 절약을 위한 모델 오프로딩

[순차적 CPU 오프로딩](#sequential_offloading)은 이전 섹션에서 설명한 것처럼 많은 메모리를 보존하지만 필요에 따라 서브모듈을 GPU로 이동하고 새 모듈이 실행될 때 즉시 CPU로 반환되기 때문에 추론 속도가 느려집니다.

전체 모델 오프로딩은 각 모델의 구성 요소인 _modules_을 처리하는 대신, 전체 모델을 GPU로 이동하는 대안입니다. 이로 인해 추론 시간에 미치는 영향은 미미하지만(파이프라인을 'cuda'로 이동하는 것과 비교하여) 여전히 약간의 메모리를 절약할 수 있습니다.

이 시나리오에서는 파이프라인의 주요 구성 요소 중 하나만(일반적으로 텍스트 인코더, unet 및 vae) GPU에 있고, 나머지는 CPU에서 대기할 것입니다.
여러 반복을 위해 실행되는 UNet과 같은 구성 요소는 더 이상 필요하지 않을 때까지 GPU에 남아 있습니다.

이 기능은 아래와 같이 파이프라인에서 `enable_model_cpu_offload()`를 호출하여 활성화할 수 있습니다.

```Python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
)

prompt = "a photo of an astronaut riding a horse on mars"
pipe.enable_model_cpu_offload()
image = pipe(prompt).images[0]
```

이는 추가적인 메모리 절약을 위한 attention slicing과도 호환됩니다.

```Python
import torch
from diffusers import StableDiffusionPipeline

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
)

prompt = "a photo of an astronaut riding a horse on mars"
pipe.enable_model_cpu_offload()
pipe.enable_attention_slicing(1)

image = pipe(prompt).images[0]
```

> [!TIP]
> 이 기능을 사용하려면 'accelerate' 버전 0.17.0 이상이 필요합니다.

## Channels Last 메모리 형식 사용하기

Channels Last 메모리 형식은 차원 순서를 보존하는 메모리에서 NCHW 텐서 배열을 대체하는 방법입니다.
Channels Last 텐서는 채널이 가장 조밀한 차원이 되는 방식으로 정렬됩니다(일명 픽셀당 이미지를 저장).
현재 모든 연산자 Channels Last 형식을 지원하는 것은 아니라 성능이 저하될 수 있으므로, 사용해보고 모델에 잘 작동하는지 확인하는 것이 좋습니다.


예를 들어 파이프라인의 UNet 모델이 channels Last 형식을 사용하도록 설정하려면 다음을 사용할 수 있습니다:

```python
print(pipe.unet.conv_out.state_dict()["weight"].stride())  # (2880, 9, 3, 1)
pipe.unet.to(memory_format=torch.channels_last)  # in-place 연산
# 2번째 차원에서 스트라이드 1을 가지는 (2880, 1, 960, 320)로, 연산이 작동함을 증명합니다.
print(pipe.unet.conv_out.state_dict()["weight"].stride())
```

## 추적(tracing)

추적은 모델을 통해 예제 입력 텐서를 통해 실행되는데, 해당 입력이 모델의 레이어를 통과할 때 호출되는 작업을 캡처하여 실행 파일 또는 'ScriptFunction'이 반환되도록 하고, 이는 just-in-time 컴파일로 최적화됩니다.

UNet 모델을 추적하기 위해 다음을 사용할 수 있습니다:

```python
import time
import torch
from diffusers import StableDiffusionPipeline
import functools

# torch 기울기 비활성화
torch.set_grad_enabled(False)

# 변수 설정
n_experiments = 2
unet_runs_per_experiment = 50


# 입력 불러오기
def generate_inputs():
    sample = torch.randn((2, 4, 64, 64), device="cuda", dtype=torch.float16)
    timestep = torch.rand(1, device="cuda", dtype=torch.float16) * 999
    encoder_hidden_states = torch.randn((2, 77, 768), device="cuda", dtype=torch.float16)
    return sample, timestep, encoder_hidden_states


pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
).to("cuda")
unet = pipe.unet
unet.eval()
unet.to(memory_format=torch.channels_last)  # Channels Last 메모리 형식 사용
unet.forward = functools.partial(unet.forward, return_dict=False)  # return_dict=False을 기본값으로 설정

# 워밍업
for _ in range(3):
    with torch.inference_mode():
        inputs = generate_inputs()
        orig_output = unet(*inputs)

# 추적
print("tracing..")
unet_traced = torch.jit.trace(unet, inputs)
unet_traced.eval()
print("done tracing")


# 워밍업 및 그래프 최적화
for _ in range(5):
    with torch.inference_mode():
        inputs = generate_inputs()
        orig_output = unet_traced(*inputs)


# 벤치마킹
with torch.inference_mode():
    for _ in range(n_experiments):
        torch.cuda.synchronize()
        start_time = time.time()
        for _ in range(unet_runs_per_experiment):
            orig_output = unet_traced(*inputs)
        torch.cuda.synchronize()
        print(f"unet traced inference took {time.time() - start_time:.2f} seconds")
    for _ in range(n_experiments):
        torch.cuda.synchronize()
        start_time = time.time()
        for _ in range(unet_runs_per_experiment):
            orig_output = unet(*inputs)
        torch.cuda.synchronize()
        print(f"unet inference took {time.time() - start_time:.2f} seconds")

# 모델 저장
unet_traced.save("unet_traced.pt")
```

그 다음, 파이프라인의 `unet` 특성을 다음과 같이 추적된 모델로 바꿀 수 있습니다.

```python
from diffusers import StableDiffusionPipeline
import torch
from dataclasses import dataclass


@dataclass
class UNet2DConditionOutput:
    sample: torch.Tensor


pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
).to("cuda")

# jitted unet 사용
unet_traced = torch.jit.load("unet_traced.pt")


# pipe.unet 삭제
class TracedUNet(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.in_channels = pipe.unet.config.in_channels
        self.device = pipe.unet.device

    def forward(self, latent_model_input, t, encoder_hidden_states):
        sample = unet_traced(latent_model_input, t, encoder_hidden_states)[0]
        return UNet2DConditionOutput(sample=sample)


pipe.unet = TracedUNet()

with torch.inference_mode():
    image = pipe([prompt] * 1, num_inference_steps=50).images[0]
```


## Memory-efficient attention

어텐션 블록의 대역폭을 최적화하는 최근 작업으로 GPU 메모리 사용량이 크게 향상되고 향상되었습니다.
@tridao의 가장 최근의 플래시 어텐션: [code](https://github.com/HazyResearch/flash-attention), [paper](https://huggingface.co/papers/2205.14135).

배치 크기 1(프롬프트 1개)의 512x512 크기로 추론을 실행할 때 몇 가지 Nvidia GPU에서 얻은 속도 향상은 다음과 같습니다:

| GPU              	| 기준 어텐션 FP16 	       | 메모리 효율적인 어텐션 FP16 	|
|------------------	|---------------------	|---------------------------------	|
| NVIDIA Tesla T4  	| 3.5it/s             	| 5.5it/s                         	|
| NVIDIA 3060 RTX  	| 4.6it/s             	| 7.8it/s                         	|
| NVIDIA A10G      	| 8.88it/s            	| 15.6it/s                        	|
| NVIDIA RTX A6000 	| 11.7it/s            	| 21.09it/s                       	|
| NVIDIA TITAN RTX  | 12.51it/s         	| 18.22it/s                       	|
| A100-SXM4-40GB    	| 18.6it/s            	| 29.it/s                        	|
| A100-SXM-80GB    	| 18.7it/s            	| 29.5it/s                        	|

이를 활용하려면 다음을 만족해야 합니다:
 - PyTorch > 1.12
 - Cuda 사용 가능
 - [xformers 라이브러리를 설치함](xformers)
```python
from diffusers import StableDiffusionPipeline
import torch

pipe = StableDiffusionPipeline.from_pretrained(
    "stable-diffusion-v1-5/stable-diffusion-v1-5",
    torch_dtype=torch.float16,
).to("cuda")

pipe.enable_xformers_memory_efficient_attention()

with torch.inference_mode():
    sample = pipe("a small cat")

# 선택: 이를 비활성화 하기 위해 다음을 사용할 수 있습니다.
# pipe.disable_xformers_memory_efficient_attention()
```


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/fp16.md" />

### Core ML로 Stable Diffusion을 실행하는 방법
https://huggingface.co/docs/diffusers/main/ko/optimization/coreml.md

# Core ML로 Stable Diffusion을 실행하는 방법

[Core ML](https://developer.apple.com/documentation/coreml)은 Apple 프레임워크에서 지원하는 모델 형식 및 머신 러닝 라이브러리입니다. macOS 또는 iOS/iPadOS 앱 내에서 Stable Diffusion 모델을 실행하는 데 관심이 있는 경우, 이 가이드에서는 기존 PyTorch 체크포인트를 Core ML 형식으로 변환하고 이를 Python 또는 Swift로 추론에 사용하는 방법을 설명합니다.

Core ML 모델은 Apple 기기에서 사용할 수 있는 모든 컴퓨팅 엔진들, 즉 CPU, GPU, Apple Neural Engine(또는 Apple Silicon Mac 및 최신 iPhone/iPad에서 사용할 수 있는 텐서 최적화 가속기인 ANE)을 활용할 수 있습니다. 모델과 실행 중인 기기에 따라 Core ML은 컴퓨팅 엔진도 혼합하여 사용할 수 있으므로, 예를 들어 모델의 일부가 CPU에서 실행되는 반면 다른 부분은 GPU에서 실행될 수 있습니다.

> [!TIP]
> PyTorch에 내장된 `mps` 가속기를 사용하여 Apple Silicon Macs에서 `diffusers` Python 코드베이스를 실행할 수도 있습니다. 이 방법은 [mps 가이드]에 자세히 설명되어 있지만 네이티브 앱과 호환되지 않습니다.

## Stable Diffusion Core ML 체크포인트

Stable Diffusion 가중치(또는 체크포인트)는 PyTorch 형식으로 저장되기 때문에 네이티브 앱에서 사용하기 위해서는 Core ML 형식으로 변환해야 합니다.

다행히도 Apple 엔지니어들이 `diffusers`를 기반으로 한 [변환 툴](https://github.com/apple/ml-stable-diffusion#-converting-models-to-core-ml)을 개발하여 PyTorch 체크포인트를 Core ML로 변환할 수 있습니다.

모델을 변환하기 전에 잠시 시간을 내어 Hugging Face Hub를 살펴보세요. 관심 있는 모델이 이미 Core ML 형식으로 제공되고 있을 가능성이 높습니다:

- [Apple](https://huggingface.co/apple) organization에는 Stable Diffusion 버전 1.4, 1.5, 2.0 base 및 2.1 base가 포함되어 있습니다.
- [coreml](https://huggingface.co/coreml) organization에는 커스텀 DreamBooth가 적용되거나, 파인튜닝된 모델이 포함되어 있습니다.
- 이 [필터](https://huggingface.co/models?pipeline_tag=text-to-image&library=coreml&p=2&sort=likes)를 사용하여 사용 가능한 모든 Core ML 체크포인트들을 반환합니다.

원하는 모델을 찾을 수 없는 경우 Apple의 [모델을 Core ML로 변환하기](https://github.com/apple/ml-stable-diffusion#-converting-models-to-core-ml) 지침을 따르는 것이 좋습니다.

## 사용할 Core ML 변형(Variant) 선택하기

Stable Diffusion 모델은 다양한 목적에 따라 다른 Core ML 변형으로 변환할 수 있습니다:

- 사용되는 어텐션 블록 유형. 어텐션 연산은 이미지 표현의 여러 영역 간의 관계에 '주의를 기울이고' 이미지와 텍스트 표현이 어떻게 연관되어 있는지 이해하는 데 사용됩니다. 어텐션 연산은 컴퓨팅 및 메모리 집약적이므로 다양한 장치의 하드웨어 특성을 고려한 다양한 구현이 존재합니다. Core ML Stable Diffusion 모델의 경우 두 가지 주의 변형이 있습니다:
    * `split_einsum` ([Apple에서 도입](https://machinelearning.apple.com/research/neural-engine-transformers)은 최신 iPhone, iPad 및 M 시리즈 컴퓨터에서 사용할 수 있는 ANE 장치에 최적화되어 있습니다.
    * "원본" 어텐션(`diffusers`에 사용되는 기본 구현)는 CPU/GPU와만 호환되며 ANE와는 호환되지 않습니다. "원본" 어텐션을 사용하여 CPU + GPU에서 모델을 실행하는 것이 ANE보다 *더* 빠를 수 있습니다. 자세한 내용은 [이 성능 벤치마크](https://huggingface.co/blog/fast-mac-diffusers#performance-benchmarks)와 커뮤니티에서 제공하는 일부 [추가 측정](https://github.com/huggingface/swift-coreml-diffusers/issues/31)을 참조하십시오.

- 지원되는 추론 프레임워크
    * `packages`는 Python 추론에 적합합니다. 네이티브 앱에 통합하기 전에 변환된 Core ML 모델을 테스트하거나, Core ML 성능을 알고 싶지만 네이티브 앱을 지원할 필요는 없는 경우에 사용할 수 있습니다. 예를 들어, 웹 UI가 있는 애플리케이션은 Python Core ML 백엔드를 완벽하게 사용할 수 있습니다.
    * Swift 코드에는 `컴파일된` 모델이 필요합니다. Hub의 `컴파일된` 모델은 iOS 및 iPadOS 기기와의 호환성을 위해 큰 UNet 모델 가중치를 여러 파일로 분할합니다. 이는 [`--chunk-unet` 변환 옵션](https://github.com/apple/ml-stable-diffusion#-converting-models-to-core-ml)에 해당합니다. 네이티브 앱을 지원하려면 `컴파일된` 변형을 선택해야 합니다.

공식 Core ML Stable Diffusion [모델](https://huggingface.co/apple/coreml-stable-diffusion-v1-4/tree/main)에는 이러한 변형이 포함되어 있지만 커뮤니티 버전은 다를 수 있습니다:

```
coreml-stable-diffusion-v1-4
├── README.md
├── original
│   ├── compiled
│   └── packages
└── split_einsum
    ├── compiled
    └── packages
```

아래와 같이 필요한 변형을 다운로드하여 사용할 수 있습니다.

## Python에서 Core ML 추론

Python에서 Core ML 추론을 실행하려면 다음 라이브러리를 설치하세요:

```bash
pip install huggingface_hub
pip install git+https://github.com/apple/ml-stable-diffusion
```

### 모델 체크포인트 다운로드하기

`컴파일된` 버전은 Swift와만 호환되므로 Python에서 추론을 실행하려면 `packages` 폴더에 저장된 버전 중 하나를 사용하세요. `원본` 또는 `split_einsum` 어텐션 중 어느 것을 사용할지 선택할 수 있습니다.

다음은 Hub에서 'models'라는 디렉토리로 'original' 어텐션 변형을 다운로드하는 방법입니다:

```Python
from huggingface_hub import snapshot_download
from pathlib import Path

repo_id = "apple/coreml-stable-diffusion-v1-4"
variant = "original/packages"

model_path = Path("./models") / (repo_id.split("/")[-1] + "_" + variant.replace("/", "_"))
snapshot_download(repo_id, allow_patterns=f"{variant}/*", local_dir=model_path, local_dir_use_symlinks=False)
print(f"Model downloaded at {model_path}")
```


### 추론[[python-inference]]

모델의 snapshot을 다운로드한 후에는 Apple의 Python 스크립트를 사용하여 테스트할 수 있습니다.

```shell
python -m python_coreml_stable_diffusion.pipeline --prompt "a photo of an astronaut riding a horse on mars" -i models/coreml-stable-diffusion-v1-4_original_packages -o </path/to/output/image> --compute-unit CPU_AND_GPU --seed 93
```

`<output-mlpackages-directory>`는 위 단계에서 다운로드한 체크포인트를 가리켜야 하며, `--compute-unit`은 추론을 허용할 하드웨어를 나타냅니다. 이는 다음 옵션 중 하나이어야 합니다: `ALL`, `CPU_AND_GPU`, `CPU_ONLY`, `CPU_AND_NE`. 선택적 출력 경로와 재현성을 위한 시드를 제공할 수도 있습니다.

추론 스크립트에서는 Stable Diffusion 모델의 원래 버전인 `CompVis/stable-diffusion-v1-4`를 사용한다고 가정합니다. 다른 모델을 사용하는 경우 추론 명령줄에서 `--model-version` 옵션을 사용하여 해당 허브 ID를 *지정*해야 합니다. 이는 이미 지원되는 모델과 사용자가 직접 학습하거나 파인튜닝한 사용자 지정 모델에 적용됩니다.

예를 들어, [`stable-diffusion-v1-5/stable-diffusion-v1-5`](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)를 사용하려는 경우입니다:

```shell
python -m python_coreml_stable_diffusion.pipeline --prompt "a photo of an astronaut riding a horse on mars" --compute-unit ALL -o output --seed 93 -i models/coreml-stable-diffusion-v1-5_original_packages --model-version stable-diffusion-v1-5/stable-diffusion-v1-5
```


## Swift에서 Core ML 추론하기

Swift에서 추론을 실행하는 것은 모델이 이미 `mlmodelc` 형식으로 컴파일되어 있기 때문에 Python보다 약간 빠릅니다. 이는 앱이 시작될 때 모델이 불러와지는 것이 눈에 띄지만, 이후 여러 번 실행하면 눈에 띄지 않을 것입니다.

### 다운로드

Mac에서 Swift에서 추론을 실행하려면 `컴파일된` 체크포인트 버전 중 하나가 필요합니다. 이전 예제와 유사하지만 `컴파일된` 변형 중 하나를 사용하여 Python 코드를 로컬로 다운로드하는 것이 좋습니다:

```Python
from huggingface_hub import snapshot_download
from pathlib import Path

repo_id = "apple/coreml-stable-diffusion-v1-4"
variant = "original/compiled"

model_path = Path("./models") / (repo_id.split("/")[-1] + "_" + variant.replace("/", "_"))
snapshot_download(repo_id, allow_patterns=f"{variant}/*", local_dir=model_path, local_dir_use_symlinks=False)
print(f"Model downloaded at {model_path}")
```

### 추론[[swift-inference]]

추론을 실행하기 위해서, Apple의 리포지토리를 복제하세요:

```bash
git clone https://github.com/apple/ml-stable-diffusion
cd ml-stable-diffusion
```

그 다음 Apple의 명령어 도구인 [Swift 패키지 관리자](https://www.swift.org/package-manager/#)를 사용합니다:

```bash
swift run StableDiffusionSample --resource-path models/coreml-stable-diffusion-v1-4_original_compiled --compute-units all "a photo of an astronaut riding a horse on mars"
```

`--resource-path`에 이전 단계에서 다운로드한 체크포인트 중 하나를 지정해야 하므로 확장자가 `.mlmodelc`인 컴파일된 Core ML 번들이 포함되어 있는지 확인하시기 바랍니다. `--compute-units`는 다음 값 중 하나이어야 합니다: `all`, `cpuOnly`, `cpuAndGPU`, `cpuAndNeuralEngine`.

자세한 내용은 [Apple의 리포지토리 안의 지침](https://github.com/apple/ml-stable-diffusion)을 참고하시기 바랍니다.


## 지원되는 Diffusers 기능

Core ML 모델과 추론 코드는 🧨 Diffusers의 많은 기능, 옵션 및 유연성을 지원하지 않습니다. 다음은 유의해야 할 몇 가지 제한 사항입니다:

- Core ML 모델은 추론에만 적합합니다. 학습이나 파인튜닝에는 사용할 수 없습니다.
- Swift에 포팅된 스케줄러는 Stable Diffusion에서 사용하는 기본 스케줄러와 `diffusers` 구현에서 Swift로 포팅한 `DPMSolverMultistepScheduler` 두 개뿐입니다. 이들 중 약 절반의 스텝으로 동일한 품질을 생성하는 `DPMSolverMultistepScheduler`를 사용하는 것이 좋습니다.
- 추론 코드에서 네거티브 프롬프트, classifier-free guidance scale 및 image-to-image 작업을 사용할 수 있습니다. depth guidance, ControlNet, latent upscalers와 같은 고급 기능은 아직 사용할 수 없습니다.

Apple의 [변환 및 추론 리포지토리](https://github.com/apple/ml-stable-diffusion)와 자체 [swift-coreml-diffusers](https://github.com/huggingface/swift-coreml-diffusers) 리포지토리는 다른 개발자들이 구축할 수 있는 기술적인 데모입니다.

누락된 기능이 있다고 생각되면 언제든지 기능을 요청하거나, 더 좋은 방법은 기여 PR을 열어주세요. :)


## 네이티브 Diffusers Swift 앱

자체 Apple 하드웨어에서 Stable Diffusion을 실행하는 쉬운 방법 중 하나는 `diffusers`와 Apple의 변환 및 추론 리포지토리를 기반으로 하는 [자체 오픈 소스 Swift 리포지토리](https://github.com/huggingface/swift-coreml-diffusers)를 사용하는 것입니다. 코드를 공부하고 [Xcode](https://developer.apple.com/xcode/)로 컴파일하여 필요에 맞게 조정할 수 있습니다. 편의를 위해 앱스토어에 [독립형 Mac 앱](https://apps.apple.com/app/diffusers/id1666309574)도 있으므로 코드나 IDE를 다루지 않고도 사용할 수 있습니다. 개발자로서 Core ML이 Stable Diffusion 앱을 구축하는 데 가장 적합한 솔루션이라고 판단했다면, 이 가이드의 나머지 부분을 사용하여 프로젝트를 시작할 수 있습니다. 여러분이 무엇을 빌드할지 기대됩니다. :)

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/coreml.md" />

### Diffusers에서의 PyTorch 2.0 가속화 지원
https://huggingface.co/docs/diffusers/main/ko/optimization/torch2.0.md

# Diffusers에서의 PyTorch 2.0 가속화 지원

`0.13.0` 버전부터 Diffusers는 [PyTorch 2.0](https://pytorch.org/get-started/pytorch-2.0/)에서의 최신 최적화를 지원합니다. 이는 다음을 포함됩니다.
1. momory-efficient attention을 사용한 가속화된 트랜스포머 지원 - `xformers`같은 추가적인 dependencies 필요 없음
2. 추가 성능 향상을 위한 개별 모델에 대한 컴파일 기능 [torch.compile](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html) 지원


## 설치
가속화된 어텐션 구현과 및 `torch.compile()`을 사용하기 위해, pip에서 최신 버전의 PyTorch 2.0을 설치되어 있고 diffusers 0.13.0. 버전 이상인지 확인하세요. 아래 설명된 바와 같이, PyTorch 2.0이 활성화되어 있을 때 diffusers는 최적화된 어텐션 프로세서([`AttnProcessor2_0`](https://github.com/huggingface/diffusers/blob/1a5797c6d4491a879ea5285c4efc377664e0332d/src/diffusers/models/attention_processor.py#L798))를 사용합니다.

```bash
pip install --upgrade torch diffusers
```

## 가속화된 트랜스포머와 `torch.compile` 사용하기.


1. **가속화된 트랜스포머 구현**

   PyTorch 2.0에는 [`torch.nn.functional.scaled_dot_product_attention`](https://pytorch.org/docs/master/generated/torch.nn.functional.scaled_dot_product_attention) 함수를 통해 최적화된 memory-efficient attention의 구현이 포함되어 있습니다. 이는 입력 및 GPU 유형에 따라 여러 최적화를 자동으로 활성화합니다. 이는 [xFormers](https://github.com/facebookresearch/xformers)의 `memory_efficient_attention`과 유사하지만 기본적으로 PyTorch에 내장되어 있습니다.

   이러한 최적화는 PyTorch 2.0이 설치되어 있고 `torch.nn.functional.scaled_dot_product_attention`을 사용할 수 있는 경우 Diffusers에서 기본적으로 활성화됩니다. 이를 사용하려면 `torch 2.0`을 설치하고 파이프라인을 사용하기만 하면 됩니다. 예를 들어:

    ```Python
    import torch
    from diffusers import DiffusionPipeline

    pipe = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16)
    pipe = pipe.to("cuda")

    prompt = "a photo of an astronaut riding a horse on mars"
    image = pipe(prompt).images[0]
    ```

    이를 명시적으로 활성화하려면(필수는 아님) 아래와 같이 수행할 수 있습니다.

    ```diff
    import torch
    from diffusers import DiffusionPipeline
    + from diffusers.models.attention_processor import AttnProcessor2_0

    pipe = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16).to("cuda")
    + pipe.unet.set_attn_processor(AttnProcessor2_0())

    prompt = "a photo of an astronaut riding a horse on mars"
    image = pipe(prompt).images[0]
    ```

    이 실행 과정은 `xFormers`만큼 빠르고 메모리적으로 효율적이어야 합니다. 자세한 내용은 [벤치마크](#benchmark)에서 확인하세요.

    파이프라인을 보다 deterministic으로 만들거나 파인 튜닝된 모델을 [Core ML](https://huggingface.co/docs/diffusers/v0.16.0/en/optimization/coreml#how-to-run-stable-diffusion-with-core-ml)과 같은 다른 형식으로 변환해야 하는 경우 바닐라 어텐션 프로세서 ([`AttnProcessor`](https://github.com/huggingface/diffusers/blob/1a5797c6d4491a879ea5285c4efc377664e0332d/src/diffusers/models/attention_processor.py#L402))로 되돌릴 수 있습니다. 일반 어텐션 프로세서를 사용하려면 `set_default_attn_processor()` 함수를 사용할 수 있습니다:

    ```Python
    import torch
    from diffusers import DiffusionPipeline
    from diffusers.models.attention_processor import AttnProcessor

    pipe = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5", torch_dtype=torch.float16).to("cuda")
    pipe.unet.set_default_attn_processor()

    prompt = "a photo of an astronaut riding a horse on mars"
    image = pipe(prompt).images[0]
    ```

2. **torch.compile**

    추가적인 속도 향상을 위해 새로운 `torch.compile` 기능을 사용할 수 있습니다. 파이프라인의 UNet은 일반적으로 계산 비용이 가장 크기 때문에 나머지 하위 모델(텍스트 인코더와 VAE)은 그대로 두고 `unet`을 `torch.compile`로 래핑합니다. 자세한 내용과 다른 옵션은 [torch 컴파일 문서](https://pytorch.org/tutorials/intermediate/torch_compile_tutorial.html)를 참조하세요.

    ```python
    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
    images = pipe(prompt, num_inference_steps=steps, num_images_per_prompt=batch_size).images
    ```

    GPU 유형에 따라 `compile()`은 가속화된 트랜스포머 최적화를 통해 **5% - 300%**의 _추가 성능 향상_을 얻을 수 있습니다. 그러나 컴파일은 Ampere(A100, 3090), Ada(4090) 및 Hopper(H100)와 같은 최신 GPU 아키텍처에서 더 많은 성능 향상을 가져올 수 있음을 참고하세요.

    컴파일은 완료하는 데 약간의 시간이 걸리므로, 파이프라인을 한 번 준비한 다음 동일한 유형의 추론 작업을 여러 번 수행해야 하는 상황에 가장 적합합니다. 다른 이미지 크기에서 컴파일된 파이프라인을 호출하면 시간적 비용이 많이 들 수 있는 컴파일 작업이 다시 트리거됩니다.


## 벤치마크

PyTorch 2.0의 효율적인 어텐션 구현과 `torch.compile`을 사용하여 가장 많이 사용되는 5개의 파이프라인에 대해 다양한 GPU와 배치 크기에 걸쳐 포괄적인 벤치마크를 수행했습니다. 여기서는 [`torch.compile()`이 최적으로 활용되도록 하는](https://github.com/huggingface/diffusers/pull/3313) `diffusers 0.17.0.dev0`을 사용했습니다.

### 벤치마킹 코드

#### Stable Diffusion text-to-image

```python
from diffusers import DiffusionPipeline
import torch

path = "stable-diffusion-v1-5/stable-diffusion-v1-5"

run_compile = True  # Set True / False

pipe = DiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
pipe = pipe.to("cuda")
pipe.unet.to(memory_format=torch.channels_last)

if run_compile:
    print("Run torch compile")
    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)

prompt = "ghibli style, a fantasy landscape with castles"

for _ in range(3):
    images = pipe(prompt=prompt).images
```

#### Stable Diffusion image-to-image

```python
from diffusers import StableDiffusionImg2ImgPipeline
import requests
import torch
from PIL import Image
from io import BytesIO

url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"

response = requests.get(url)
init_image = Image.open(BytesIO(response.content)).convert("RGB")
init_image = init_image.resize((512, 512))

path = "stable-diffusion-v1-5/stable-diffusion-v1-5"

run_compile = True  # Set True / False

pipe = StableDiffusionImg2ImgPipeline.from_pretrained(path, torch_dtype=torch.float16)
pipe = pipe.to("cuda")
pipe.unet.to(memory_format=torch.channels_last)

if run_compile:
    print("Run torch compile")
    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)

prompt = "ghibli style, a fantasy landscape with castles"

for _ in range(3):
    image = pipe(prompt=prompt, image=init_image).images[0]
```

#### Stable Diffusion - inpainting

```python
from diffusers import StableDiffusionInpaintPipeline
import requests
import torch
from PIL import Image
from io import BytesIO

url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"

def download_image(url):
    response = requests.get(url)
    return Image.open(BytesIO(response.content)).convert("RGB")


img_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo.png"
mask_url = "https://raw.githubusercontent.com/CompVis/latent-diffusion/main/data/inpainting_examples/overture-creations-5sI6fQgYIuo_mask.png"

init_image = download_image(img_url).resize((512, 512))
mask_image = download_image(mask_url).resize((512, 512))

path = "runwayml/stable-diffusion-inpainting"

run_compile = True  # Set True / False

pipe = StableDiffusionInpaintPipeline.from_pretrained(path, torch_dtype=torch.float16)
pipe = pipe.to("cuda")
pipe.unet.to(memory_format=torch.channels_last)

if run_compile:
    print("Run torch compile")
    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)

prompt = "ghibli style, a fantasy landscape with castles"

for _ in range(3):
    image = pipe(prompt=prompt, image=init_image, mask_image=mask_image).images[0]
```

#### ControlNet

```python
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
import requests
import torch
from PIL import Image
from io import BytesIO

url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"

response = requests.get(url)
init_image = Image.open(BytesIO(response.content)).convert("RGB")
init_image = init_image.resize((512, 512))

path = "stable-diffusion-v1-5/stable-diffusion-v1-5"

run_compile = True  # Set True / False
controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    path, controlnet=controlnet, torch_dtype=torch.float16
)

pipe = pipe.to("cuda")
pipe.unet.to(memory_format=torch.channels_last)
pipe.controlnet.to(memory_format=torch.channels_last)

if run_compile:
    print("Run torch compile")
    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
    pipe.controlnet = torch.compile(pipe.controlnet, mode="reduce-overhead", fullgraph=True)

prompt = "ghibli style, a fantasy landscape with castles"

for _ in range(3):
    image = pipe(prompt=prompt, image=init_image).images[0]
```

#### IF text-to-image + upscaling

```python
from diffusers import DiffusionPipeline
import torch

run_compile = True  # Set True / False

pipe = DiffusionPipeline.from_pretrained("DeepFloyd/IF-I-M-v1.0", variant="fp16", text_encoder=None, torch_dtype=torch.float16)
pipe.to("cuda")
pipe_2 = DiffusionPipeline.from_pretrained("DeepFloyd/IF-II-M-v1.0", variant="fp16", text_encoder=None, torch_dtype=torch.float16)
pipe_2.to("cuda")
pipe_3 = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-x4-upscaler", torch_dtype=torch.float16)
pipe_3.to("cuda")


pipe.unet.to(memory_format=torch.channels_last)
pipe_2.unet.to(memory_format=torch.channels_last)
pipe_3.unet.to(memory_format=torch.channels_last)

if run_compile:
    pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
    pipe_2.unet = torch.compile(pipe_2.unet, mode="reduce-overhead", fullgraph=True)
    pipe_3.unet = torch.compile(pipe_3.unet, mode="reduce-overhead", fullgraph=True)

prompt = "the blue hulk"

prompt_embeds = torch.randn((1, 2, 4096), dtype=torch.float16)
neg_prompt_embeds = torch.randn((1, 2, 4096), dtype=torch.float16)

for _ in range(3):
    image = pipe(prompt_embeds=prompt_embeds, negative_prompt_embeds=neg_prompt_embeds, output_type="pt").images
    image_2 = pipe_2(image=image, prompt_embeds=prompt_embeds, negative_prompt_embeds=neg_prompt_embeds, output_type="pt").images
    image_3 = pipe_3(prompt=prompt, image=image, noise_level=100).images
```

PyTorch 2.0 및 `torch.compile()`로 얻을 수 있는 가능한 속도 향상에 대해, [Stable Diffusion text-to-image pipeline](StableDiffusionPipeline)에 대한 상대적인 속도 향상을 보여주는 차트를 5개의 서로 다른 GPU 제품군(배치 크기 4)에 대해 나타냅니다:

![t2i_speedup](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/pt2_benchmarks/t2i_speedup.png)

To give you an even better idea of how this speed-up holds for the other pipelines presented above, consider the following
plot that shows the benchmarking numbers from an A100 across three different batch sizes
(with PyTorch 2.0 nightly and `torch.compile()`):
이 속도 향상이 위에 제시된 다른 파이프라인에 대해서도 어떻게 유지되는지 더 잘 이해하기 위해, 세 가지의 다른 배치 크기에 걸쳐 A100의 벤치마킹(PyTorch 2.0 nightly 및 `torch.compile() 사용) 수치를 보여주는 차트를 보입니다:

![a100_numbers](https://huggingface.co/datasets/diffusers/docs-images/resolve/main/pt2_benchmarks/a100_numbers.png)

_(위 차트의 벤치마크 메트릭은 **초당 iteration 수(iterations/second)**입니다)_

그러나 투명성을 위해 모든 벤치마킹 수치를 공개합니다!

다음 표들에서는, **_초당 처리되는 iteration_** 수 측면에서의 결과를 보여줍니다.

### A100 (batch size: 1)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 21.66 | 23.13 | 44.03 | 49.74 |
| SD - img2img | 21.81 | 22.40 | 43.92 | 46.32 |
| SD - inpaint | 22.24 | 23.23 | 43.76 | 49.25 |
| SD - controlnet | 15.02 | 15.82 | 32.13 | 36.08 |
| IF | 20.21 / <br>13.84 / <br>24.00 | 20.12 / <br>13.70 / <br>24.03 | ❌ | 97.34 / <br>27.23 / <br>111.66 |

### A100 (batch size: 4)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 11.6 | 13.12 | 14.62 | 17.27 |
| SD - img2img | 11.47 | 13.06 | 14.66 | 17.25 |
| SD - inpaint | 11.67 | 13.31 | 14.88 | 17.48 |
| SD - controlnet | 8.28 | 9.38 | 10.51 | 12.41 |
| IF | 25.02 | 18.04 | ❌ | 48.47 |

### A100 (batch size: 16)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 3.04 | 3.6 | 3.83 | 4.68 |
| SD - img2img | 2.98 | 3.58 | 3.83 | 4.67 |
| SD - inpaint | 3.04 | 3.66 | 3.9 | 4.76 |
| SD - controlnet | 2.15 | 2.58 | 2.74 | 3.35 |
| IF | 8.78 | 9.82 | ❌ | 16.77 |

### V100 (batch size: 1)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 18.99 | 19.14 | 20.95 | 22.17 |
| SD - img2img | 18.56 | 19.18 | 20.95 | 22.11 |
| SD - inpaint | 19.14 | 19.06 | 21.08 | 22.20 |
| SD - controlnet | 13.48 | 13.93 | 15.18 | 15.88 |
| IF |  20.01 / <br>9.08 / <br>23.34 | 19.79 / <br>8.98 / <br>24.10 | ❌ | 55.75 / <br>11.57 / <br>57.67 |

### V100 (batch size: 4)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 5.96 | 5.89 | 6.83 | 6.86 |
| SD - img2img | 5.90 | 5.91 | 6.81 | 6.82 |
| SD - inpaint | 5.99 | 6.03 | 6.93 | 6.95 |
| SD - controlnet | 4.26 | 4.29 | 4.92 | 4.93 |
| IF | 15.41 | 14.76 | ❌ | 22.95 |

### V100 (batch size: 16)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 1.66 | 1.66 | 1.92 | 1.90 |
| SD - img2img | 1.65 | 1.65 | 1.91 | 1.89 |
| SD - inpaint | 1.69 | 1.69 | 1.95 | 1.93 |
| SD - controlnet | 1.19 | 1.19 | OOM after warmup | 1.36 |
| IF | 5.43 | 5.29 | ❌ | 7.06 |

### T4 (batch size: 1)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 6.9 | 6.95 | 7.3 | 7.56 |
| SD - img2img | 6.84 | 6.99 | 7.04 | 7.55 |
| SD - inpaint | 6.91 | 6.7 | 7.01 | 7.37 |
| SD - controlnet | 4.89 | 4.86 | 5.35 | 5.48 |
| IF | 17.42 / <br>2.47 / <br>18.52 | 16.96 / <br>2.45 / <br>18.69 | ❌ | 24.63 / <br>2.47 / <br>23.39 |

### T4 (batch size: 4)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 1.79 | 1.79 | 2.03 | 1.99 |
| SD - img2img | 1.77 | 1.77 | 2.05 | 2.04 |
| SD - inpaint | 1.81 | 1.82 | 2.09 | 2.09 |
| SD - controlnet | 1.34 | 1.27 | 1.47 | 1.46 |
| IF | 5.79 |  5.61 | ❌ | 7.39 |

### T4 (batch size: 16)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 2.34s | 2.30s | OOM after 2nd iteration | 1.99s |
| SD - img2img | 2.35s | 2.31s | OOM after warmup | 2.00s |
| SD - inpaint | 2.30s | 2.26s | OOM after 2nd iteration | 1.95s |
| SD - controlnet | OOM after 2nd iteration | OOM after 2nd iteration | OOM after warmup | OOM after warmup |
| IF * | 1.44 | 1.44 | ❌ | 1.94 |

### RTX 3090 (batch size: 1)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 22.56 | 22.84 | 23.84 | 25.69 |
| SD - img2img | 22.25 | 22.61 | 24.1 | 25.83 |
| SD - inpaint | 22.22 | 22.54 | 24.26 | 26.02 |
| SD - controlnet | 16.03 | 16.33 | 17.38 | 18.56 |
| IF | 27.08 / <br>9.07 / <br>31.23 | 26.75 / <br>8.92 / <br>31.47 | ❌ | 68.08 / <br>11.16 / <br>65.29 |

### RTX 3090 (batch size: 4)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 6.46 | 6.35 | 7.29 | 7.3 |
| SD - img2img | 6.33 | 6.27 | 7.31 | 7.26 |
| SD - inpaint | 6.47 | 6.4 | 7.44 | 7.39 |
| SD - controlnet | 4.59 | 4.54 | 5.27 | 5.26 |
| IF | 16.81 | 16.62 | ❌ | 21.57 |

### RTX 3090 (batch size: 16)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 1.7 | 1.69 | 1.93 | 1.91 |
| SD - img2img | 1.68 | 1.67 | 1.93 | 1.9 |
| SD - inpaint | 1.72 | 1.71 | 1.97 | 1.94 |
| SD - controlnet | 1.23 | 1.22 | 1.4 | 1.38 |
| IF | 5.01 | 5.00 | ❌ | 6.33 |

### RTX 4090 (batch size: 1)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 40.5 | 41.89 | 44.65 | 49.81 |
| SD - img2img | 40.39 | 41.95 | 44.46 | 49.8 |
| SD - inpaint | 40.51 | 41.88 | 44.58 | 49.72 |
| SD - controlnet | 29.27 | 30.29 | 32.26 | 36.03 |
| IF | 69.71 / <br>18.78 / <br>85.49 | 69.13 / <br>18.80 / <br>85.56 | ❌ | 124.60 / <br>26.37 / <br>138.79 |

### RTX 4090 (batch size: 4)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 12.62 | 12.84 | 15.32 | 15.59 |
| SD - img2img | 12.61 | 12,.79 | 15.35 | 15.66 |
| SD - inpaint | 12.65 | 12.81 | 15.3 | 15.58 |
| SD - controlnet | 9.1 | 9.25 | 11.03 | 11.22 |
| IF | 31.88 | 31.14 | ❌ | 43.92 |

### RTX 4090 (batch size: 16)

| **Pipeline** | **torch 2.0 - <br>no compile** | **torch nightly - <br>no compile** | **torch 2.0 - <br>compile** | **torch nightly - <br>compile** |
|:---:|:---:|:---:|:---:|:---:|
| SD - txt2img | 3.17 | 3.2 | 3.84 | 3.85 |
| SD - img2img | 3.16 | 3.2 | 3.84 | 3.85 |
| SD - inpaint | 3.17 | 3.2 | 3.85 | 3.85 |
| SD - controlnet | 2.23 | 2.3 | 2.7 | 2.75 |
| IF | 9.26 | 9.2 | ❌ | 13.31 |

## 참고

* Follow [this PR](https://github.com/huggingface/diffusers/pull/3313) for more details on the environment used for conducting the benchmarks.
* For the IF pipeline and batch sizes > 1, we only used a batch size of >1 in the first IF pipeline for text-to-image generation and NOT for upscaling. So, that means the two upscaling pipelines received a batch size of 1.

*Thanks to [Horace He](https://github.com/Chillee) from the PyTorch team for their support in improving our support of `torch.compile()` in Diffusers.*

* 벤치마크 수행에 사용된 환경에 대한 자세한 내용은 [이 PR](https://github.com/huggingface/diffusers/pull/3313)을 참조하세요.
* IF 파이프라인와 배치 크기 > 1의 경우 첫 번째 IF 파이프라인에서 text-to-image 생성을 위한 배치 크기 > 1만 사용했으며 업스케일링에는 사용하지 않았습니다. 즉, 두 개의 업스케일링 파이프라인이 배치 크기 1임을 의미합니다.

*Diffusers에서 `torch.compile()` 지원을 개선하는 데 도움을 준 PyTorch 팀의 [Horace He](https://github.com/Chillee)에게 감사드립니다.*

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/torch2.0.md" />

### Apple Silicon (M1/M2)에서 Stable Diffusion을 사용하는 방법
https://huggingface.co/docs/diffusers/main/ko/optimization/mps.md

# Apple Silicon (M1/M2)에서 Stable Diffusion을 사용하는 방법

Diffusers는 Stable Diffusion 추론을 위해 PyTorch `mps`를 사용해 Apple 실리콘과 호환됩니다. 다음은 Stable Diffusion이 있는 M1 또는 M2 컴퓨터를 사용하기 위해 따라야 하는 단계입니다.

## 요구 사항

- Apple silicon (M1/M2) 하드웨어의 Mac 컴퓨터.
- macOS 12.6 또는 이후 (13.0 또는 이후 추천).
- Python arm64 버전
- PyTorch 2.0(추천) 또는 1.13(`mps`를 지원하는 최소 버전). Yhttps://pytorch.org/get-started/locally/의 지침에 따라 `pip` 또는 `conda`로 설치할 수 있습니다.


## 추론 파이프라인

아래 코도는 익숙한 `to()` 인터페이스를 사용하여 `mps` 백엔드로 Stable Diffusion 파이프라인을 M1 또는 M2 장치로 이동하는 방법을 보여줍니다.


> [!WARNING]
> **PyTorch 1.13을 사용 중일 때 ** 추가 일회성 전달을 사용하여 파이프라인을 "프라이밍"하는 것을 추천합니다. 이것은 발견한 이상한 문제에 대한 임시 해결 방법입니다. 첫 번째 추론 전달은 후속 전달와 약간 다른 결과를 생성합니다. 이 전달은 한 번만 수행하면 되며 추론 단계를 한 번만 사용하고 결과를 폐기해도 됩니다.

이전 팁에서 설명한 것들을 포함한 여러 문제를 해결하므로 PyTorch 2 이상을 사용하는 것이 좋습니다.


```python
# `hf auth login`에 로그인되어 있음을 확인
from diffusers import DiffusionPipeline

pipe = DiffusionPipeline.from_pretrained("stable-diffusion-v1-5/stable-diffusion-v1-5")
pipe = pipe.to("mps")

# 컴퓨터가 64GB 이하의 RAM 램일 때 추천
pipe.enable_attention_slicing()

prompt = "a photo of an astronaut riding a horse on mars"

# 처음 "워밍업" 전달 (위 설명을 보세요)
_ = pipe(prompt, num_inference_steps=1)

# 결과는 워밍업 전달 후의 CPU 장치의 결과와 일치합니다.
image = pipe(prompt).images[0]
```

## 성능 추천

M1/M2 성능은 메모리 압력에 매우 민감합니다. 시스템은 필요한 경우 자동으로 스왑되지만 스왑할 때 성능이 크게 저하됩니다.


특히 컴퓨터의 시스템 RAM이 64GB 미만이거나 512 × 512픽셀보다 큰 비표준 해상도에서 이미지를 생성하는 경우, 추론 중에 메모리 압력을 줄이고 스와핑을 방지하기 위해 *어텐션 슬라이싱*을 사용하는 것이 좋습니다. 어텐션 슬라이싱은 비용이 많이 드는 어텐션 작업을 한 번에 모두 수행하는 대신 여러 단계로 수행합니다. 일반적으로 범용 메모리가 없는 컴퓨터에서 ~20%의 성능 영향을 미치지만 64GB 이상이 아닌 경우 대부분의 Apple Silicon 컴퓨터에서 *더 나은 성능*이 관찰되었습니다.

```python
pipeline.enable_attention_slicing()
```

## Known Issues

- 여러 프롬프트를 배치로 생성하는 것은 [충돌이 발생하거나 안정적으로 작동하지 않습니다](https://github.com/huggingface/diffusers/issues/363). 우리는 이것이 [PyTorch의 `mps` 백엔드](https://github.com/pytorch/pytorch/issues/84039)와 관련이 있다고 생각합니다. 이 문제는 해결되고 있지만 지금은 배치 대신 반복 방법을 사용하는 것이 좋습니다.

<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/mps.md" />

### 추론을 위해 ONNX 런타임을 사용하는 방법
https://huggingface.co/docs/diffusers/main/ko/optimization/onnx.md

# 추론을 위해 ONNX 런타임을 사용하는 방법

🤗 Diffusers는 ONNX Runtime과 호환되는 Stable Diffusion 파이프라인을 제공합니다. 이를 통해 ONNX(CPU 포함)를 지원하고 PyTorch의 가속 버전을 사용할 수 없는 모든 하드웨어에서 Stable Diffusion을 실행할 수 있습니다.

## 설치

다음 명령어로 ONNX Runtime를 지원하는 🤗 Optimum를 설치합니다:

```sh
pip install optimum["onnxruntime"]
```

## Stable Diffusion 추론

아래 코드는 ONNX 런타임을 사용하는 방법을 보여줍니다. `StableDiffusionPipeline` 대신 `OnnxStableDiffusionPipeline`을 사용해야 합니다.
PyTorch 모델을 불러오고 즉시 ONNX 형식으로 변환하려는 경우 `export=True`로 설정합니다.

```python
from optimum.onnxruntime import ORTStableDiffusionPipeline

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipe = ORTStableDiffusionPipeline.from_pretrained(model_id, export=True)
prompt = "a photo of an astronaut riding a horse on mars"
images = pipe(prompt).images[0]
pipe.save_pretrained("./onnx-stable-diffusion-v1-5")
```

파이프라인을 ONNX 형식으로 오프라인으로 내보내고 나중에 추론에 사용하려는 경우,
[`optimum-cli export`](https://huggingface.co/docs/optimum/main/en/exporters/onnx/usage_guides/export_a_model#exporting-a-model-to-onnx-using-the-cli) 명령어를 사용할 수 있습니다:

```bash
optimum-cli export onnx --model stable-diffusion-v1-5/stable-diffusion-v1-5 sd_v15_onnx/
```

그 다음 추론을 수행합니다:

```python
from optimum.onnxruntime import ORTStableDiffusionPipeline

model_id = "sd_v15_onnx"
pipe = ORTStableDiffusionPipeline.from_pretrained(model_id)
prompt = "a photo of an astronaut riding a horse on mars"
images = pipe(prompt).images[0]
```

Notice that we didn't have to specify `export=True` above.

[Optimum 문서](https://huggingface.co/docs/optimum/)에서 더 많은 예시를 찾을 수 있습니다.

## 알려진 이슈들

- 여러 프롬프트를 배치로 생성하면 너무 많은 메모리가 사용되는 것 같습니다. 이를 조사하는 동안, 배치 대신 반복 방법이 필요할 수도 있습니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/onnx.md" />

### 추론을 위한 OpenVINO 사용 방법
https://huggingface.co/docs/diffusers/main/ko/optimization/open_vino.md

# 추론을 위한 OpenVINO 사용 방법

🤗 [Optimum](https://github.com/huggingface/optimum-intel)은 OpenVINO와 호환되는 Stable Diffusion 파이프라인을 제공합니다.
이제 다양한 Intel 프로세서에서 OpenVINO Runtime으로 쉽게 추론을 수행할 수 있습니다. ([여기](https://docs.openvino.ai/latest/openvino_docs_OV_UG_supported_plugins_Supported_Devices.html)서 지원되는 전 기기 목록을 확인하세요).

## 설치

다음 명령어로 🤗 Optimum을 설치합니다:

```sh
pip install optimum["openvino"]
```

## Stable Diffusion 추론

OpenVINO 모델을 불러오고 OpenVINO 런타임으로 추론을 실행하려면 `StableDiffusionPipeline`을 `OVStableDiffusionPipeline`으로 교체해야 합니다. PyTorch 모델을 불러오고 즉시 OpenVINO 형식으로 변환하려는 경우 `export=True`로 설정합니다.

```python
from optimum.intel.openvino import OVStableDiffusionPipeline

model_id = "stable-diffusion-v1-5/stable-diffusion-v1-5"
pipe = OVStableDiffusionPipeline.from_pretrained(model_id, export=True)
prompt = "a photo of an astronaut riding a horse on mars"
images = pipe(prompt).images[0]
```

[Optimum 문서](https://huggingface.co/docs/optimum/intel/inference#export-and-inference-of-stable-diffusion-models)에서 (정적 reshaping과 모델 컴파일 등의) 더 많은 예시들을 찾을 수 있습니다.


<EditOnGithub source="https://github.com/huggingface/diffusers/blob/main/docs/source/ko/optimization/open_vino.md" />
