app.py

# app.py
from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
import torch
from diffusers import AutoPipelineForText2Image
from contextlib import asynccontextmanager
import io
import base64
import os

# --- Pydantic Models ---
class ImageRequest(BaseModel):
    prompt: str
    negative_prompt: str = ""
    steps: int = 20 # Lowered for faster CPU inference

class ImageResponse(BaseModel):
    image_base64: str

# --- App State and Lifespan ---
app_state = {}

@asynccontextmanager
async def lifespan(app: FastAPI):
    # Load the model on startup
    hf_token = os.getenv("HF_TOKEN") # Still needed to accept terms
    if not hf_token:
        raise RuntimeError("HF_TOKEN environment variable not set! Please add it in the Space settings.")

    # --- *** THESE ARE THE CHANGES FOR CPU *** ---
    # 1. Use the smaller Stable Diffusion v1.5 model
    model_id = "runwayml/stable-diffusion-v1-5"
    
    print(f"Loading model: {model_id} for CPU...")
    
    # 2. Load the pipeline without GPU-specific settings
    pipe = AutoPipelineForText2Image.from_pretrained(
        model_id,
        token=hf_token
    )
    # Note: We do not use .to("cuda")
    # --- ************************************** ---
    
    app_state["pipe"] = pipe
    print("Model loaded successfully onto CPU.")
    yield
    # Clean up on shutdown
    app_state.clear()
    print("Resources cleaned up.")

# --- FastAPI App ---
app = FastAPI(lifespan=lifespan)

@app.get("/")
def root():
    return {"status": "Text-to-Image CPU API is running"}

@app.post("/generate-image", response_model=ImageResponse)
def generate_image(request: ImageRequest):
    if "pipe" not in app_state:
        raise HTTPException(status_code=503, detail="Model is not ready.")

    pipe = app_state["pipe"]
    
    print(f"Generating image for prompt: '{request.prompt}'")
    try:
        image = pipe(
            prompt=request.prompt,
            negative_prompt=request.negative_prompt,
            num_inference_steps=request.steps
        ).images[0]

        buffer = io.BytesIO()
        image.save(buffer, format="PNG")
        img_str = base64.b64encode(buffer.getvalue()).decode("utf-8")

        return ImageResponse(image_base64=img_str)

    except Exception as e:
        print(f"Error during image generation: {e}")
        raise HTTPException(status_code=500, detail=str(e))