app.py

import os
import sys
import subprocess
import argparse
from pathlib import Path
import torch
import datetime
import numpy as np
from PIL import Image
import imageio
import spaces

# --- Part 1: Auto-Setup (Clone Repo & Download Weights) ---

REPO_URL = "https://github.com/Tencent-Hunyuan/HunyuanVideo-1.5.git"
REPO_DIR = "HunyuanVideo-1.5"
MODEL_DIR = "ckpts"
HF_REPO_ID = "tencent/HunyuanVideo"

# Configuration
TRANSFORMER_VERSION = "480p_i2v_distilled"
DTYPE = torch.bfloat16
# Set to False if you have >40GB VRAM and want everything on GPU constantly.
# Set to True (Default) to allow running on 16GB-24GB cards via CPU offloading.
ENABLE_OFFLOADING = True 

def setup_environment():
    """Clones the repo and downloads weights if they don't exist."""
    print("=" * 50)
    print("Checking Environment & Dependencies...")
    
    # 1. Clone Repository
    if not os.path.exists(REPO_DIR):
        print(f"Cloning repository from {REPO_URL}...")
        subprocess.run(["git", "clone", REPO_URL], check=True)
    else:
        print(f"Repository {REPO_DIR} exists.")

    # 2. Add Repo to Python Path
    repo_path = os.path.abspath(REPO_DIR)
    if repo_path not in sys.path:
        sys.path.insert(0, repo_path)

    # 3. Download Weights
    if not os.path.exists(MODEL_DIR) or not os.listdir(MODEL_DIR):
        print(f"Downloading weights from {HF_REPO_ID} to {MODEL_DIR}...")
        try:
            from huggingface_hub import snapshot_download
            allow_patterns = [
                f"transformer/{TRANSFORMER_VERSION}/*",
                "vae/*",
                "text_encoder/*",
                "vision_encoder/*",
                "scheduler/*",
                "tokenizer/*"
            ]
            snapshot_download(repo_id=HF_REPO_ID, local_dir=MODEL_DIR, allow_patterns=allow_patterns)
            print("Download complete.")
        except Exception as e:
            print(f"Error downloading weights: {e}")
            sys.exit(1)
    print("Environment Ready.")
    print("=" * 50)

# Run setup immediately
setup_environment()

# --- Part 2: Imports from Cloned Repo ---

# Set Env Vars for HyVideo
if 'PYTORCH_CUDA_ALLOC_CONF' not in os.environ:
    os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'expandable_segments:True'
os.environ['RANK'] = '0'
os.environ['WORLD_SIZE'] = '1'

try:
    from hyvideo.pipelines.hunyuan_video_pipeline import HunyuanVideo_1_5_Pipeline
    from hyvideo.commons.parallel_states import initialize_parallel_state
    from hyvideo.commons.infer_state import initialize_infer_state
except ImportError as e:
    print(f"CRITICAL ERROR: Could not import hyvideo modules. {e}")
    sys.exit(1)

import gradio as gr

# --- Part 3: Model Initialization (Pre-Load) ---

# Initialize Distributed/Infer States
#parallel_dims = initialize_parallel_state(sp=1)
#if torch.cuda.is_available():
#    torch.cuda.set_device(0)

class ArgsNamespace:
    def __init__(self):
        self.use_sageattn = False
        self.sage_blocks_range = "0-53"
        self.enable_torch_compile = False

initialize_infer_state(ArgsNamespace())

# Global Pipeline Variable
pipe = None

def pre_load_model():
    """Loads the model into memory/GPU before UI launch."""
    global pipe
    print(f"⏳ Initializing Pipeline ({TRANSFORMER_VERSION})... this may take a moment...")
    
    try:
        pipe = HunyuanVideo_1_5_Pipeline.create_pipeline(
            pretrained_model_name_or_path=MODEL_DIR,
            transformer_version=TRANSFORMER_VERSION,
            enable_offloading=ENABLE_OFFLOADING,
            enable_group_offloading=ENABLE_OFFLOADING,
            transformer_dtype=DTYPE,
        )
        print("✅ Model loaded successfully!")
        if not ENABLE_OFFLOADING:
            print("   Model is fully resident on GPU.")
        else:
            print("   Model loaded with CPU Offloading enabled (optimizes VRAM usage).")
    except Exception as e:
        print(f"❌ Failed to load model: {e}")
        sys.exit(1)

def save_video_tensor(video_tensor, path, fps=24):
    if isinstance(video_tensor, list): video_tensor = video_tensor[0]
    if video_tensor.ndim == 5: video_tensor = video_tensor[0]
    vid = (video_tensor * 255).clamp(0, 255).to(torch.uint8)
    vid = vid.permute(1, 2, 3, 0).cpu().numpy()
    imageio.mimwrite(path, vid, fps=fps)

@spaces.GPU(duration=120)
def generate(input_image, prompt, length, steps, shift, seed, guidance):
    if pipe is None:
        raise gr.Error("Pipeline not initialized!")
    
    if input_image is None:
        raise gr.Error("Reference image required.")

    if isinstance(input_image, np.ndarray):
        input_image = Image.fromarray(input_image).convert("RGB")

    if seed == -1: seed = torch.randint(0, 1000000, (1,)).item()
    generator = torch.Generator(device="cpu").manual_seed(int(seed))

    print(f"Generating: {prompt} | Seed: {seed}")

    try:
        output = pipe(
            prompt=prompt,
            height=480, width=854, aspect_ratio="16:9",
            video_length=int(length),
            num_inference_steps=int(steps),
            guidance_scale=float(guidance),
            flow_shift=float(shift),
            reference_image=input_image,
            seed=int(seed),
            generator=generator,
            output_type="pt",
            enable_sr=False,
            return_dict=True
        )
    except Exception as e:
        raise gr.Error(f"Inference Failed: {e}")

    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
    os.makedirs("outputs", exist_ok=True)
    output_path = f"outputs/gen_{timestamp}.mp4"
    save_video_tensor(output.videos, output_path)
    
    return output_path

# --- Part 4: UI Definition & Launch ---

def create_ui():
    with gr.Blocks(title="HunyuanVideo 1.5 I2V") as demo:
        gr.Markdown(f"### 🎬 HunyuanVideo 1.5 I2V ({TRANSFORMER_VERSION})")
        gr.Markdown("Model is pre-loaded. Ready to generate.")
        
        with gr.Row():
            with gr.Column():
                img = gr.Image(label="Reference", type="pil", height=250)
                prompt = gr.Textbox(label="Prompt", placeholder="Describe motion...", lines=2)
                with gr.Row():
                    steps = gr.Slider(2, 20, value=6, step=1, label="Steps")
                    guidance = gr.Slider(1.0, 5.0, value=1.0, step=0.1, label="Guidance")
                with gr.Row():
                    shift = gr.Slider(1.0, 20.0, value=5.0, step=0.5, label="Shift")
                    length = gr.Slider(1, 129, value=61, step=4, label="Length")
                    seed = gr.Number(value=-1, label="Seed", precision=0)
                btn = gr.Button("Generate", variant="primary")
            
            with gr.Column():
                out = gr.Video(label="Result", autoplay=True)
        
        btn.click(generate, inputs=[img, prompt, length, steps, shift, seed, guidance], outputs=[out])
    return demo

if __name__ == "__main__":
    # 1. Execute the pre-load BEFORE the UI launches
    pre_load_model()
    
    # 2. Launch UI
    ui = create_ui()
    ui.queue().launch(server_name="0.0.0.0", share=True)