app.py

import gradio as gr
import torch
from diffusers import OnnxStableDiffusionPipeline
import os

# Use CPU optimized model
os.environ["HF_HOME"] = "/tmp/hf_cache"

# Using a proper ONNX model for CPU inference
MODEL_ID = "runwayml/stable-diffusion-v1-5"

print(f"Loading ONNX model {MODEL_ID}...")

try:
    # Use ONNX pipeline for better CPU performance
    pipe = OnnxStableDiffusionPipeline.from_pretrained(
        MODEL_ID,
        revision="onnx",
        provider="CPUExecutionProvider",
        torch_dtype=torch.float32,
        low_cpu_mem_usage=True
    )
    pipe.safety_checker = None  # Disable safety checker for simplicity
    pipe.set_progress_bar_config(disable=True)
    print("ONNX Model loaded successfully!")
except Exception as e:
    print(f"Failed to load ONNX model: {e}")
    print("Falling back to regular Stable Diffusion pipeline...")
    from diffusers import StableDiffusionPipeline
    pipe = StableDiffusionPipeline.from_pretrained(
        MODEL_ID,
        torch_dtype=torch.float32,
        low_cpu_mem_usage=True
    )
    pipe = pipe.to("cpu")
    pipe.safety_checker = None
    pipe.set_progress_bar_config(disable=True)
    print("Model loaded!")

def generate(prompt):
    try:
        result = pipe(
            prompt=prompt,
            num_inference_steps=40,
            num_images_per_prompt=1,
            guidance_scale=7.5
        )
        return result.images[0]
    except Exception as e:
        print(f"Error during generation: {e}")
        return None

interface = gr.Interface(
    fn=generate,
    inputs=gr.Textbox(lines=3, label="Prompt", placeholder="Enter your image prompt here..."),
    outputs=gr.Image(type="pil", label="Generated Image"),
    title="⚡ ONNX Stable Diffusion",
    description="Faster CPU-based image generation with ONNX optimization"
)

interface.launch()