app.py

import os
import torch
from PIL import Image
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel
import gradio as gr

# Disable oneDNN custom operations
os.environ['TF_ENABLE_ONEDNN_OPTS'] = '0'

# Clear PyTorch cache
torch.cuda.empty_cache()

# Check if CUDA is available
device = "cuda" if torch.cuda.is_available() else "cpu"
if device == "cuda":
    print("CUDA is available. Device count:", torch.cuda.device_count())
    print("Current device:", torch.cuda.current_device())
    print("Device name:", torch.cuda.get_device_name(torch.cuda.current_device()))
else:
    print("CUDA is not available. Using CPU.")

# Load ControlNet model with OpenPose pre-trained weights from Hugging Face
controlnet = ControlNetModel.from_pretrained("lllyasviel/control_v11p_sd15_openpose", torch_dtype=torch.float16)

# Load the Stable Diffusion model
pipe = StableDiffusionControlNetPipeline.from_pretrained(
    "runwayml/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
).to(device)

# Function for inference
def generate_image(prompt, target_image, pose_image):
    try:
        # Resize images
        target_image = target_image.resize((512, 512))
        pose_image = pose_image.resize((512, 512))

        # Generate image with ControlNet
        output = pipe(prompt=prompt, image=target_image, control_image=pose_image, num_inference_steps=50)

        # Return the result
        return output["sample"][0]
    except Exception as e:
        print(f"Error during image generation: {e}")
        return None

# Setup Gradio Interface
interface = gr.Interface(
    fn=generate_image,
    inputs=[
        gr.Textbox(label="Prompt"),
        gr.Image(label="Target Image", type="pil"),
        gr.Image(label="Pose Image (Reference)", type="pil")
    ],
    outputs=gr.Image(label="Generated Image")
)

# Launch the interface
interface.launch()