app.py

import gradio as gr
import torch
from diffusers import StableDiffusionPipeline
from PIL import Image
import traceback
from typing import Optional

# Stable Diffusion模型相关设置
model_id: str = "runwayml/stable-diffusion-v1-5"
device: str = "cpu"  # force CPU usage for compatibility

image_generator_pipe: Optional[StableDiffusionPipeline] = None

try:
    pipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float32)
    image_generator_pipe = pipe.to(device)
except Exception as e:
    print(f"Failed to load Stable Diffusion model: {e}")

# 提示词优化函数（简单版）
def optimize_prompt_simple(short_description: str) -> str:
    optimized_prompt = f"Generate a high-quality, detailed image based on the following description: {short_description}"
    return optimized_prompt

# 图像生成函数
def generate_image_sd(short_description: str,
                      negative_prompt: str,
                      guidance_scale: float,
                      num_inference_steps: int) -> Image.Image:
    optimized_prompt = optimize_prompt_simple(short_description)

    try:
        with torch.no_grad():
            if image_generator_pipe is None:
                raise RuntimeError("Stable Diffusion pipeline is not available.")

            output = image_generator_pipe(
                prompt=optimized_prompt,
                negative_prompt=negative_prompt,
                guidance_scale=guidance_scale,
                num_inference_steps=num_inference_steps
            )
            image = output.images[0] if output.images else None

        if not image:
            raise RuntimeError("No image was returned from the generation pipeline.")

        return image
    except Exception as e:
        traceback.print_exc()
        raise gr.Error(f"Image generation failed: {str(e)}")
    # Gradio界面
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    with gr.Row():
        with gr.Column(scale=1):
            short_description = gr.Textbox(label="Short Description", placeholder="A magical treehouse in the sky")
            optimized_prompt_display = gr.Textbox(label="Optimized Prompt", interactive=False)
            neg_prompt = gr.Textbox(label="Negative Prompt", placeholder="blurry, distorted, watermark")
            guidance = gr.Slider(1.0, 15.0, value=7.5, step=0.5, label="Guidance Scale")
            steps = gr.Slider(10, 50, value=25, step=1, label="Inference Steps")
            generate_btn = gr.Button("Generate Image")

        with gr.Column(scale=1):
            output_image = gr.Image(label="Generated Image", type="pil")

    # 当用户输入简短描述时，自动优化提示词并显示
    short_description.input(
        fn=lambda x: optimize_prompt_simple(x),
        inputs=short_description,
        outputs=optimized_prompt_display
    )

    generate_btn.click(
        fn=generate_image_sd,
        inputs=[short_description, neg_prompt, guidance, steps],
        outputs=output_image
    )

if __name__ == "__main__":
    if not image_generator_pipe:
        print("WARNING: Stable Diffusion pipeline is not available. UI will launch, but generation will fail.")

    demo.launch(server_name="0.0.0.0", server_port=7860)