app.py

import gradio as gr
import torch
from diffusers import StableDiffusionXLPipeline, StableDiffusionXLImg2ImgPipeline
from PIL import Image
import os
import gc
import time
from typing import Optional, Tuple
from huggingface_hub import hf_hub_download
import requests

# Global pipeline variables
txt2img_pipe = None
img2img_pipe = None
device = "cuda" if torch.cuda.is_available() else "cpu"

# Hugging Face model configuration
MODEL_REPO = "ajsbsd/CyberRealistic-Pony"
MODEL_FILENAME = "cyberrealisticPony_v110.safetensors"
LOCAL_MODEL_PATH = "./models/cyberrealisticPony_v110.safetensors"

def clear_memory():
    """Clear GPU memory"""
    if torch.cuda.is_available():
        torch.cuda.empty_cache()
    gc.collect()

def download_model():
    """Download model from Hugging Face if not already cached"""
    try:
        # Create models directory if it doesn't exist
        os.makedirs("./models", exist_ok=True)
        
        # Check if model already exists locally
        if os.path.exists(LOCAL_MODEL_PATH):
            print(f"Model already exists at {LOCAL_MODEL_PATH}")
            return LOCAL_MODEL_PATH
        
        print(f"Downloading model from {MODEL_REPO}/{MODEL_FILENAME}...")
        print("This may take a while on first run...")
        
        # Download the model file
        model_path = hf_hub_download(
            repo_id=MODEL_REPO,
            filename=MODEL_FILENAME,
            local_dir="./models",
            local_dir_use_symlinks=False,
            resume_download=True
        )
        
        print(f"Model downloaded successfully to {model_path}")
        return model_path
        
    except Exception as e:
        print(f"Error downloading model: {e}")
        print("Attempting to use cached version or fallback...")
        
        # Try to use Hugging Face cache directly
        try:
            cached_path = hf_hub_download(
                repo_id=MODEL_REPO,
                filename=MODEL_FILENAME,
                resume_download=True
            )
            print(f"Using cached model at {cached_path}")
            return cached_path
        except Exception as cache_error:
            print(f"Cache fallback failed: {cache_error}")
            return None

def load_models():
    """Load both text2img and img2img pipelines with Hugging Face integration"""
    global txt2img_pipe, img2img_pipe
    
    # Download model if needed
    model_path = download_model()
    
    if model_path is None:
        print("Failed to download or locate model file")
        return None, None
    
    if not os.path.exists(model_path):
        print(f"Model file not found after download: {model_path}")
        return None, None
    
    if txt2img_pipe is None:
        try:
            print("Loading CyberRealistic Pony Text2Img model...")
            txt2img_pipe = StableDiffusionXLPipeline.from_single_file(
                model_path,
                torch_dtype=torch.float16 if device == "cuda" else torch.float32,
                use_safetensors=True,
                variant="fp16" if device == "cuda" else None
            )
            
            # Memory optimizations
            txt2img_pipe.enable_attention_slicing()
            
            if device == "cuda":
                try:
                    txt2img_pipe.enable_model_cpu_offload()
                    print("Text2Img CPU offload enabled")
                except Exception as e:
                    print(f"Text2Img CPU offload failed: {e}")
                    txt2img_pipe = txt2img_pipe.to(device)
            else:
                txt2img_pipe = txt2img_pipe.to(device)
                    
            print("Text2Img model loaded successfully!")
            
        except Exception as e:
            print(f"Error loading Text2Img model: {e}")
            return None, None
    
    if img2img_pipe is None:
        try:
            print("Loading CyberRealistic Pony Img2Img model...")
            img2img_pipe = StableDiffusionXLImg2ImgPipeline.from_single_file(
                model_path,
                torch_dtype=torch.float16 if device == "cuda" else torch.float32,
                use_safetensors=True,
                variant="fp16" if device == "cuda" else None
            )
            
            # Memory optimizations
            img2img_pipe.enable_attention_slicing()
            
            if device == "cuda":
                try:
                    img2img_pipe.enable_model_cpu_offload()
                    print("Img2Img CPU offload enabled")
                except Exception as e:
                    print(f"Img2Img CPU offload failed: {e}")
                    img2img_pipe = img2img_pipe.to(device)
            else:
                img2img_pipe = img2img_pipe.to(device)
                    
            print("Img2Img model loaded successfully!")
            
        except Exception as e:
            print(f"Error loading Img2Img model: {e}")
            return txt2img_pipe, None
    
    return txt2img_pipe, img2img_pipe

def enhance_prompt(prompt: str, add_quality_tags: bool = True) -> str:
    """Enhance prompt with Pony-style tags"""
    if not prompt.strip():
        return prompt
        
    # Don't add tags if already present
    if prompt.startswith("score_") or not add_quality_tags:
        return prompt
        
    quality_tags = "score_9, score_8_up, score_7_up, masterpiece, best quality, highly detailed"
    return f"{quality_tags}, {prompt}"

def validate_dimensions(width: int, height: int) -> Tuple[int, int]:
    """Ensure dimensions are valid for SDXL"""
    # SDXL works best with dimensions divisible by 64
    width = ((width + 63) // 64) * 64
    height = ((height + 63) // 64) * 64
    
    # Ensure reasonable limits
    width = max(512, min(1536, width))
    height = max(512, min(1536, height))
    
    return width, height

def generate_txt2img(prompt, negative_prompt, num_steps, guidance_scale, width, height, seed, add_quality_tags):
    """Generate image from text prompt with enhanced error handling"""
    global txt2img_pipe
    
    if not prompt.strip():
        return None, "Please enter a prompt"
        
    # Load models if not already loaded
    if txt2img_pipe is None:
        txt2img_pipe, _ = load_models()
        if txt2img_pipe is None:
            return None, "Failed to load Text2Img model. Please check your internet connection and try again."
    
    try:
        # Clear memory before generation
        clear_memory()
        
        # Validate and fix dimensions
        width, height = validate_dimensions(width, height)
        
        # Set seed for reproducibility
        generator = None
        if seed != -1:
            generator = torch.Generator(device=device).manual_seed(int(seed))
        
        # Enhance prompt
        enhanced_prompt = enhance_prompt(prompt, add_quality_tags)
        
        print(f"Generating with prompt: {enhanced_prompt[:100]}...")
        start_time = time.time()
        
        # Generate image
        with torch.no_grad():
            result = txt2img_pipe(
                prompt=enhanced_prompt,
                negative_prompt=negative_prompt or "",
                num_inference_steps=int(num_steps),
                guidance_scale=float(guidance_scale),
                width=width,
                height=height,
                generator=generator
            )
        
        generation_time = time.time() - start_time
        status = f"Text2Img: Generated successfully in {generation_time:.1f}s (Size: {width}x{height})"
        
        return result.images[0], status
        
    except Exception as e:
        error_msg = f"Text2Img generation failed: {str(e)}"
        print(error_msg)
        return None, error_msg
    finally:
        clear_memory()

def generate_img2img(input_image, prompt, negative_prompt, num_steps, guidance_scale, strength, seed, add_quality_tags):
    """Generate image from input image + text prompt with enhanced error handling"""
    global img2img_pipe
    
    if input_image is None:
        return None, "Please upload an input image for Img2Img"
    
    if not prompt.strip():
        return None, "Please enter a prompt"
        
    # Load models if not already loaded
    if img2img_pipe is None:
        _, img2img_pipe = load_models()
        if img2img_pipe is None:
            return None, "Failed to load Img2Img model. Please check your internet connection and try again."
    
    try:
        # Clear memory before generation
        clear_memory()
        
        # Set seed for reproducibility
        generator = None
        if seed != -1:
            generator = torch.Generator(device=device).manual_seed(int(seed))
        
        # Enhance prompt
        enhanced_prompt = enhance_prompt(prompt, add_quality_tags)
        
        # Process input image
        if isinstance(input_image, Image.Image):
            # Ensure RGB format
            if input_image.mode != 'RGB':
                input_image = input_image.convert('RGB')
                
            # Resize to reasonable dimensions while maintaining aspect ratio
            original_size = input_image.size
            max_size = 1024
            input_image.thumbnail((max_size, max_size), Image.Resampling.LANCZOS)
            
            # Ensure dimensions are divisible by 64
            w, h = input_image.size
            w, h = validate_dimensions(w, h)
            input_image = input_image.resize((w, h), Image.Resampling.LANCZOS)
        
        print(f"Generating with prompt: {enhanced_prompt[:100]}...")
        start_time = time.time()
        
        # Generate image
        with torch.no_grad():
            result = img2img_pipe(
                prompt=enhanced_prompt,
                negative_prompt=negative_prompt or "",
                image=input_image,
                num_inference_steps=int(num_steps),
                guidance_scale=float(guidance_scale),
                strength=float(strength),
                generator=generator
            )
        
        generation_time = time.time() - start_time
        status = f"Img2Img: Generated successfully in {generation_time:.1f}s (Strength: {strength})"
        
        return result.images[0], status
        
    except Exception as e:
        error_msg = f"Img2Img generation failed: {str(e)}"
        print(error_msg)
        return None, error_msg
    finally:
        clear_memory()

# Default negative prompt (improved)
DEFAULT_NEGATIVE = """
(low quality:1.4), (worst quality:1.4), (bad quality:1.3), (normal quality:1.2), lowres, jpeg artifacts, blurry, noisy, ugly, deformed, disfigured, malformed, poorly drawn, bad art, amateur, render, 3D, cgi, 
(text, signature, watermark, username, copyright:1.5), 
(extra limbs:1.5), (missing limbs:1.5), (extra fingers:1.5), (missing fingers:1.5), (mutated hands:1.5), (bad hands:1.4), (poorly drawn hands:1.3), (ugly hands:1.2), 
(bad anatomy:1.4), (deformed body:1.3), (unnatural body:1.2), (cross-eyed:1.3), (skewed eyes:1.3), (imperfect eyes:1.2), (ugly eyes:1.2), (asymmetrical face:1.2), (unnatural face:1.2), 
(blush:1.1), (shadow on skin:1.1), (shaded skin:1.1), (dark skin:1.1), 
abstract, simplified, unrealistic, impressionistic, cartoon, anime, drawing, sketch, illustration, painting, censored, grayscale, monochrome, out of frame, cropped, distorted.
"""

# Create Gradio interface with enhanced styling
with gr.Blocks(
    title="CyberRealistic Pony Image Generator", 
    theme=gr.themes.Soft(),
    css="""
    .gradio-container {
        max-width: 1200px !important;
    }
    .tab-nav button {
        font-size: 16px;
        font-weight: bold;
    }
    """
) as demo:
    gr.Markdown("""
    # 🎨 CyberRealistic Pony Image Generator (Hugging Face Edition)
    
    Generate high-quality images using the CyberRealistic Pony SDXL model from Hugging Face.
    
    **Features:**
    - 🎨 Text-to-Image generation
    - 🖼️ Image-to-Image transformation
    - 🎯 Automatic quality tag enhancement
    - ⚡ Memory optimized for better performance
    - 🤗 Auto-downloads model from Hugging Face
    
    **Note:** On first run, the model will be downloaded from Hugging Face (this may take a few minutes).
    """)
    
    with gr.Tabs():
        # Text2Image Tab
        with gr.TabItem("🎨 Text to Image"):
            with gr.Row():
                with gr.Column(scale=1):
                    # Input controls for Text2Img
                    txt2img_prompt = gr.Textbox(
                        label="Prompt",
                        placeholder="Enter your image description...",
                        value="beautiful landscape with mountains and lake at sunset",
                        lines=3
                    )
                    
                    txt2img_negative = gr.Textbox(
                        label="Negative Prompt",
                        value=DEFAULT_NEGATIVE,
                        lines=3
                    )
                    
                    txt2img_quality_tags = gr.Checkbox(
                        label="Add Quality Tags",
                        value=True
                    )
                    
                    with gr.Row():
                        txt2img_steps = gr.Slider(
                            minimum=10,
                            maximum=50,
                            value=25,
                            step=1,
                            label="Inference Steps"
                        )
                        
                        txt2img_guidance = gr.Slider(
                            minimum=1.0,
                            maximum=20.0,
                            value=7.5,
                            step=0.5,
                            label="Guidance Scale"
                        )
                    
                    with gr.Row():
                        txt2img_width = gr.Slider(
                            minimum=512,
                            maximum=1536,
                            value=1024,
                            step=64,
                            label="Width"
                        )
                        
                        txt2img_height = gr.Slider(
                            minimum=512,
                            maximum=1536,
                            value=1024,
                            step=64,
                            label="Height"
                        )
                    
                    txt2img_seed = gr.Number(
                        label="Seed (-1 for random)",
                        value=-1,
                        precision=0
                    )
                    
                    txt2img_btn = gr.Button("🎨 Generate Image", variant="primary")
                    
                with gr.Column(scale=2):
                    # Output for Text2Img
                    txt2img_output = gr.Image(
                        label="Generated Image", 
                        type="pil",
                        height=600
                    )
                    txt2img_status = gr.Textbox(label="Status", interactive=False)
        
        # Image2Image Tab
        with gr.TabItem("🖼️ Image to Image"):
            with gr.Row():
                with gr.Column(scale=1):
                    # Input controls for Img2Img
                    img2img_input = gr.Image(
                        label="Input Image",
                        type="pil",
                        height=300
                    )
                    
                    img2img_prompt = gr.Textbox(
                        label="Prompt",
                        placeholder="Describe how to modify the image...",
                        value="in the style of a digital painting, vibrant colors",
                        lines=3
                    )
                    
                    img2img_negative = gr.Textbox(
                        label="Negative Prompt",
                        value=DEFAULT_NEGATIVE,
                        lines=3
                    )
                    
                    img2img_quality_tags = gr.Checkbox(
                        label="Add Quality Tags",
                        value=True
                    )
                    
                    with gr.Row():
                        img2img_steps = gr.Slider(
                            minimum=10,
                            maximum=50,
                            value=25,
                            step=1,
                            label="Inference Steps"
                        )
                        
                        img2img_guidance = gr.Slider(
                            minimum=1.0,
                            maximum=20.0,
                            value=7.5,
                            step=0.5,
                            label="Guidance Scale"
                        )
                    
                    img2img_strength = gr.Slider(
                        minimum=0.1,
                        maximum=1.0,
                        value=0.75,
                        step=0.05,
                        label="Denoising Strength (Lower = more like input, Higher = more creative)"
                    )
                    
                    img2img_seed = gr.Number(
                        label="Seed (-1 for random)",
                        value=-1,
                        precision=0
                    )
                    
                    img2img_btn = gr.Button("🖼️ Transform Image", variant="primary")
                    
                with gr.Column(scale=2):
                    # Output for Img2Img
                    img2img_output = gr.Image(
                        label="Generated Image", 
                        type="pil",
                        height=600
                    )
                    img2img_status = gr.Textbox(label="Status", interactive=False)
    
    # Event handlers
    txt2img_btn.click(
        fn=generate_txt2img,
        inputs=[txt2img_prompt, txt2img_negative, txt2img_steps, txt2img_guidance, 
                txt2img_width, txt2img_height, txt2img_seed, txt2img_quality_tags],
        outputs=[txt2img_output, txt2img_status]
    )
    
    img2img_btn.click(
        fn=generate_img2img,
        inputs=[img2img_input, img2img_prompt, img2img_negative, txt2img_steps, img2img_guidance, 
                img2img_strength, img2img_seed, img2img_quality_tags],
        outputs=[img2img_output, img2img_status]
    )

# Load models on startup
print("Initializing CyberRealistic Pony Generator (Hugging Face Edition)...")
print(f"Device: {device}")
print(f"Model Repository: {MODEL_REPO}")
print(f"Model File: {MODEL_FILENAME}")

# Pre-load models in a separate thread to avoid blocking startup
import threading

def preload_models():
    """Pre-load models in background"""
    try:
        print("Starting background model loading...")
        load_models()
        print("Background model loading completed!")
    except Exception as e:
        print(f"Background model loading failed: {e}")

# Start background loading
loading_thread = threading.Thread(target=preload_models, daemon=True)
loading_thread.start()

if __name__ == "__main__":
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,
        show_error=True
    )