local_server.py

import os
import gradio as gr
import torch
from PIL import Image
import numpy as np
from models.ledits_model import LEDITSModel
from utils.image_processing import preprocess_image, postprocess_image
from utils.feature_detection import detect_features, create_mask

# Initialize models with GPU support
def initialize_models(device=None):
    if device is None:
        device = "cuda" if torch.cuda.is_available() else "cpu"
    
    print(f"Using device: {device}")
    if device == "cuda":
        print(f"GPU: {torch.cuda.get_device_name(0)}")
        print(f"Memory: {torch.cuda.get_device_properties(0).total_memory / 1024**3:.2f} GB")
    
    ledits_model = LEDITSModel(device=device)
    return ledits_model

# Main editing function
def edit_image(image, feature_type, modification_type, intensity, 
               num_inference_steps, guidance_scale, resolution,
               custom_prompt="", use_custom_prompt=False):
    if image is None:
        return None, "Please upload an image first."
    
    try:
        # Convert to numpy array if needed
        if isinstance(image, Image.Image):
            image_np = np.array(image)
        else:
            image_np = image
        
        # Resize image based on resolution setting
        if resolution != "Original":
            max_dim = int(resolution.split("x")[0])
            height, width = image_np.shape[:2]
            
            if height > width:
                new_height = min(max_dim, height)
                new_width = int(width * (new_height / height))
            else:
                new_width = min(max_dim, width)
                new_height = int(height * (new_width / width))
            
            image_np = Image.fromarray(image_np).resize((new_width, new_height), Image.LANCZOS)
            image_np = np.array(image_np)
            
        # Preprocess image
        processed_image = preprocess_image(image_np)
        
        # Detect features and create mask
        features = detect_features(processed_image)
        mask = create_mask(processed_image, feature_type, features)
        
        # Get model
        ledits_model = initialize_models()
        
        # Prepare prompt
        if use_custom_prompt and custom_prompt:
            prompt = custom_prompt
        else:
            prompt = f"{feature_type} {modification_type}"
        
        # Apply edit with custom parameters
        edited_image = ledits_model.edit_image(
            processed_image, 
            mask, 
            prompt, 
            intensity=intensity,
            guidance_scale=guidance_scale,
            num_inference_steps=num_inference_steps
        )
        
        # Postprocess
        final_image = postprocess_image(edited_image, processed_image, mask)
        
        return final_image, "Edit completed successfully."
    
    except Exception as e:
        import traceback
        traceback.print_exc()
        return image, f"Error during editing: {str(e)}"

# Create the server interface
def create_server():
    with gr.Blocks(title="AI-Powered Facial & Body Feature Editor - Local Server") as server:
        gr.Markdown("# AI-Powered Facial & Body Feature Editor - Local GPU Server")
        gr.Markdown("This is the local GPU server component. Keep this running to process edits from the Hugging Face Space.")
        
        # Input components
        with gr.Row():
            with gr.Column():
                input_image = gr.Image(label="Input Image", type="pil")
                feature_type = gr.Dropdown(
                    choices=["Eyes", "Nose", "Lips", "Face Shape", "Hair", "Body"],
                    label="Feature Type",
                    value="Eyes"
                )
                modification_type = gr.Dropdown(
                    choices=["Larger", "Smaller", "Change Color", "Change Shape"],
                    label="Modification Type",
                    value="Larger"
                )
                intensity = gr.Slider(
                    minimum=0.1,
                    maximum=1.0,
                    value=0.5,
                    step=0.1,
                    label="Intensity"
                )
                custom_prompt = gr.Textbox(
                    label="Custom Prompt",
                    placeholder="e.g., blue eyes with long eyelashes"
                )
                use_custom_prompt = gr.Checkbox(
                    label="Use Custom Prompt",
                    value=False
                )
                
                # Performance settings
                with gr.Group():
                    gr.Markdown("### Performance Settings")
                    num_inference_steps = gr.Slider(
                        minimum=5,
                        maximum=50,
                        value=20,
                        step=1,
                        label="Inference Steps (lower = faster, higher = better quality)"
                    )
                    guidance_scale = gr.Slider(
                        minimum=1.0,
                        maximum=15.0,
                        value=7.5,
                        step=0.5,
                        label="Guidance Scale (lower = more creative, higher = more accurate)"
                    )
                    resolution = gr.Dropdown(
                        choices=["Original", "512x512", "768x768", "1024x1024"],
                        label="Processing Resolution",
                        value="512x512"
                    )
            
            with gr.Column():
                output_image = gr.Image(label="Output Image", type="pil")
                status_text = gr.Textbox(label="Status", interactive=False)
                
                # GPU info display
                gpu_info = gr.Textbox(label="GPU Information", interactive=False)
                
                def get_gpu_info():
                    if torch.cuda.is_available():
                        return f"GPU: {torch.cuda.get_device_name(0)}\nVRAM: {torch.cuda.get_device_properties(0).total_memory / 1024**3:.2f} GB\nCUDA Version: {torch.version.cuda}"
                    else:
                        return "No GPU detected. Using CPU mode."
                
                # Display GPU info on load
                gpu_info.value = get_gpu_info()
        
        # Create the API endpoint
        edit_image_api = gr.Interface(
            fn=edit_image,
            inputs=[
                input_image,
                feature_type,
                modification_type,
                intensity,
                num_inference_steps,
                guidance_scale,
                resolution,
                custom_prompt,
                use_custom_prompt
            ],
            outputs=[output_image, status_text],
            title="AI-Powered Facial & Body Feature Editor API",
            description="Local GPU server API endpoint",
            allow_flagging="never"
        )
    
    return server

# Launch the server
if __name__ == "__main__":
    # Check for GPU
    if torch.cuda.is_available():
        print(f"GPU detected: {torch.cuda.get_device_name(0)}")
        print(f"VRAM: {torch.cuda.get_device_properties(0).total_memory / 1024**3:.2f} GB")
    else:
        print("No GPU detected. Using CPU mode.")
    
    # Create and launch server
    server = create_server()
    server.queue()
    server.launch(server_name="0.0.0.0", share=True)