app.py

import spaces
import os
import gc
import gradio as gr
import torch
import json
import utils
import logging
from diffusers.models import AutoencoderKL
from diffusers import StableDiffusionXLImg2ImgPipeline
from config import (
    MODELS,
    MIN_IMAGE_SIZE,
    MAX_IMAGE_SIZE,
    OUTPUT_DIR,
    DEFAULT_NEGATIVE_PROMPT,
    DEFAULT_ASPECT_RATIO,
    QUALITY_TAGS,
    examples,
    sampler_list,
    aspect_ratios,
    css
)
import time
from typing import List, Dict, Tuple

# Enhanced logging configuration
logging.basicConfig(
    level=logging.INFO,
    format='%(asctime)s - %(name)s - %(levelname)s - %(message)s',
    datefmt='%Y-%m-%d %H:%M:%S'
)
logger = logging.getLogger(__name__)

# Constants
IS_COLAB = utils.is_google_colab() or os.getenv("IS_COLAB") == "1"
HF_TOKEN = os.getenv("HF_TOKEN")
CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES") == "1" 

# PyTorch settings for better performance and determinism
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
torch.backends.cuda.matmul.allow_tf32 = True

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
logger.info(f"Using device: {device}")

class GenerationError(Exception):
    """Custom exception for generation errors"""
    pass

def validate_prompt(prompt: str) -> str:
    """Validate and clean up the input prompt."""
    if not isinstance(prompt, str):
        raise GenerationError("Prompt must be a string")
    try:
        # Ensure proper UTF-8 encoding/decoding
        prompt = prompt.encode('utf-8').decode('utf-8')
        # Add space between ! and ,
        prompt = prompt.replace("!,", "! ,")
    except UnicodeError:
        raise GenerationError("Invalid characters in prompt")
    
    # Only check if the prompt is completely empty or only whitespace
    if not prompt or prompt.isspace():
        raise GenerationError("Prompt cannot be empty")
    return prompt.strip()

def validate_dimensions(width: int, height: int) -> None:
    """Validate image dimensions."""
    if not MIN_IMAGE_SIZE <= width <= MAX_IMAGE_SIZE:
        raise GenerationError(f"Width must be between {MIN_IMAGE_SIZE} and {MAX_IMAGE_SIZE}")
        
    if not MIN_IMAGE_SIZE <= height <= MAX_IMAGE_SIZE:
        raise GenerationError(f"Height must be between {MIN_IMAGE_SIZE} and {MAX_IMAGE_SIZE}")

@spaces.GPU
def generate(
    prompt: str,
    negative_prompt: str = DEFAULT_NEGATIVE_PROMPT,
    seed: int = 0,
    custom_width: int = 1024,
    custom_height: int = 1024,
    guidance_scale: float = 6.0,
    num_inference_steps: int = 25,
    sampler: str = "Euler a",
    model_name: str = "v14",
    aspect_ratio_selector: str = DEFAULT_ASPECT_RATIO,
    use_upscaler: bool = False,
    upscaler_strength: float = 0.55,
    upscale_by: float = 1.5,
    add_quality_tags: bool = True,
    progress: gr.Progress = gr.Progress(track_tqdm=True),
) -> Tuple[List[str], Dict]:
    """Generate images based on the given parameters."""
    start_time = time.time()
    upscaler_pipe = None
    backup_scheduler = None
    
    try:
        # Memory management
        torch.cuda.empty_cache()
        gc.collect()

        # Input validation
        prompt = validate_prompt(prompt)
        if negative_prompt:
            negative_prompt = negative_prompt.encode('utf-8').decode('utf-8')
        
        validate_dimensions(custom_width, custom_height)
        
        # Set up generation
        generator = utils.seed_everything(seed)
        width, height = utils.aspect_ratio_handler(
            aspect_ratio_selector,
            custom_width,
            custom_height,
        )

        # Process prompts
        if add_quality_tags:
            prompt = QUALITY_TAGS.format(prompt=prompt)

        prompt, negative_prompt = utils.preprocess_prompt(
            prompt, negative_prompt
        )    

        width, height = utils.preprocess_image_dimensions(width, height)

        # Set up pipeline
        pipe = pipes[model_name]
        backup_scheduler = pipe.scheduler
        pipe.scheduler = utils.get_scheduler(pipe.scheduler.config, sampler)

        if use_upscaler:
            upscaler_pipe = StableDiffusionXLImg2ImgPipeline(**pipe.components)
            
        # Prepare metadata
        metadata = {
            "prompt": prompt,
            "negative_prompt": negative_prompt,
            "resolution": f"{width} x {height}",
            "guidance_scale": guidance_scale,
            "num_inference_steps": num_inference_steps,
            "seed": seed,
            "sampler": sampler,
            "Model": "Ikastrious " + model_name
        }

        if use_upscaler:
            new_width = int(width * upscale_by)
            new_height = int(height * upscale_by)
            metadata["use_upscaler"] = {
                "upscale_method": "nearest-exact",
                "upscaler_strength": upscaler_strength,
                "upscale_by": upscale_by,
                "new_resolution": f"{new_width} x {new_height}",
            }
        else:
            metadata["use_upscaler"] = None
        
        logger.info(f"Starting generation with parameters: {json.dumps(metadata, indent=4)}")

        # Generate images
        if use_upscaler:
            latents = pipe(
                prompt=prompt,
                negative_prompt=negative_prompt,
                width=width,
                height=height,
                guidance_scale=guidance_scale,
                num_inference_steps=num_inference_steps,
                generator=generator,
                output_type="latent",
            ).images
            upscaled_latents = utils.upscale(latents, "nearest-exact", upscale_by)
            images = upscaler_pipe(
                prompt=prompt,
                negative_prompt=negative_prompt,
                image=upscaled_latents,
                guidance_scale=guidance_scale,
                num_inference_steps=num_inference_steps,
                strength=upscaler_strength,
                generator=generator,
                output_type="pil",
            ).images
        else:
            images = pipe(
                prompt=prompt,
                negative_prompt=negative_prompt,
                width=width,
                height=height,
                guidance_scale=guidance_scale,
                num_inference_steps=num_inference_steps,
                generator=generator,
                output_type="pil",
            ).images

        # Save images
        if images:
            total = len(images)
            image_paths = []
            for idx, image in enumerate(images, 1):
                progress(idx/total, desc="Saving images...")
                path = utils.save_image(image, metadata, OUTPUT_DIR, IS_COLAB)
                image_paths.append(path)
                logger.info(f"Image {idx}/{total} saved as {path}")

        generation_time = time.time() - start_time
        logger.info(f"Generation completed successfully in {generation_time:.2f} seconds")
        metadata["generation_time"] = f"{generation_time:.2f}s"
        
        return image_paths, metadata

    except GenerationError as e:
        logger.warning(f"Generation validation error: {str(e)}")
        raise gr.Error(str(e))
    except Exception as e:
        logger.exception("Unexpected error during generation")
        raise gr.Error(f"Generation failed: {str(e)}")
    finally:
        # Cleanup
        torch.cuda.empty_cache()
        gc.collect()
        
        if upscaler_pipe is not None:
            del upscaler_pipe
        
        if backup_scheduler is not None and pipe is not None:
            pipe.scheduler = backup_scheduler
            
        utils.free_memory()

# Model initialization
if torch.cuda.is_available():
    pipes = {}

    try:
        logger.info("Loading VAE and pipeline...")
        vae = AutoencoderKL.from_pretrained(
            "madebyollin/sdxl-vae-fp16-fix",
            torch_dtype=torch.float16,
        )

        for model in MODELS:
            pipes[model['name']] = utils.load_pipeline(model['path'], device, vae=vae)
            logger.info(f"Pipeline for {model} loaded successfully on GPU!")
        logger.info("Pipeline loaded successfully on GPU!")
    except Exception as e:
        logger.error(f"Error loading VAE, falling back to default: {e}")

        for model in MODELS:
            if model['name'] not in pipes:
                pipes[model['name']] = utils.load_pipeline(model['path'], device)
                logger.info(f"Pipeline for {model} loaded successfully on GPU!")
else:
    logger.warning("CUDA not available, running on CPU")
    # check if os.environ keys have VSCODE, if not, load the model on CPU
    skip = False
    for key in dict(os.environ).keys():
        if "VSCODE" in key:
            skip = True
            break

    if not skip:
        logger.info("Loading pipeline on CPU...")
        pipe = utils.load_pipeline(MODELS[0]['name'], torch.device("cpu"))
        logger.info("Pipeline loaded successfully on CPU!")


with gr.Blocks(css=css, theme="Nymbo/Nymbo_Theme_5") as demo:
    gr.HTML(
        """
        <div class="header">
            <div class="title">Ikastrious</div>
            <div class="subtitle">Gradio demo for <a href="https://civitai.com/models/874216" target="_blank">Ikastrious</a></div>
            <span class="thanks">Thanks to <a href="https://huggingface.co/Asahina2K">Asahina2K</a> for the space code base</span>
        </div>
        """,
    )
    
    with gr.Row():
        with gr.Column(scale=2):
            with gr.Group():
                prompt = gr.Text(
                    label="Prompt",
                    max_lines=5,
                    placeholder="Describe what you want to generate",
                    info="Enter your image generation prompt here. Be specific and descriptive for better results.",
                )
                negative_prompt = gr.Text(
                    label="Negative Prompt",
                    max_lines=5,
                    placeholder="Describe what you want to avoid",
                    value=DEFAULT_NEGATIVE_PROMPT,
                    info="Specify elements you don't want in the image.",
                )
                add_quality_tags = gr.Checkbox(
                    label="Quality Tags",
                    value=True,
                    info="Add quality-enhancing tags to your prompt automatically.",
                )
                model_name = gr.Radio(
                        label="Model",
                        choices=[model['name'] for model in MODELS],
                        value=MODELS[0]['name'],
                        container=True,
                        info="Select the model to use for image generation.",
                    )
            with gr.Accordion(label="More Settings", open=False):
                with gr.Group():
                    aspect_ratio_selector = gr.Radio(
                        label="Aspect Ratio",
                        choices=aspect_ratios,
                        value=DEFAULT_ASPECT_RATIO,
                        container=True,
                        info="Choose the dimensions of your image.",
                    )
                with gr.Group(visible=False) as custom_resolution:
                    with gr.Row():
                        custom_width = gr.Slider(
                            label="Width",
                            minimum=MIN_IMAGE_SIZE,
                            maximum=MAX_IMAGE_SIZE,
                            step=8,
                            value=1536,
                            info=f"Image width (must be between {MIN_IMAGE_SIZE} and {MAX_IMAGE_SIZE})",
                        )
                        custom_height = gr.Slider(
                            label="Height",
                            minimum=MIN_IMAGE_SIZE,
                            maximum=MAX_IMAGE_SIZE,
                            step=8,
                            value=1536,
                            info=f"Image height (must be between {MIN_IMAGE_SIZE} and {MAX_IMAGE_SIZE})",
                        )
                with gr.Group():
                    use_upscaler = gr.Checkbox(
                        label="Use Upscaler",
                        value=False,
                        info="Enable high-resolution upscaling.",
                    )
                    with gr.Row() as upscaler_row:
                        upscaler_strength = gr.Slider(
                            label="Strength",
                            minimum=0,
                            maximum=1,
                            step=0.05,
                            value=0.55,
                            visible=False,
                            info="Control how much the upscaler affects the final image.",
                        )
                        upscale_by = gr.Slider(
                            label="Upscale by",
                            minimum=1,
                            maximum=1.5,
                            step=0.1,
                            value=1.5,
                            visible=False,
                            info="Multiplier for the final image resolution.",
                        )
                with gr.Accordion(label="Advanced Parameters", open=False):
                    with gr.Group():
                        sampler = gr.Dropdown(
                            label="Sampler",
                            choices=sampler_list,
                            interactive=True,
                            value="Euler a",
                            info="Different samplers can produce varying results.",
                        )
                    with gr.Group():
                        seed = gr.Slider(
                            label="Seed",
                            minimum=0,
                            maximum=utils.MAX_SEED,
                            step=1,
                            value=0,
                            info="Set a specific seed for reproducible results.",
                        )
                        randomize_seed = gr.Checkbox(
                            label="Randomize seed",
                            value=True,
                            info="Generate a new random seed for each image.",
                        )
                    with gr.Group():
                        with gr.Row():
                            guidance_scale = gr.Slider(
                                label="Guidance scale",
                                minimum=1,
                                maximum=12,
                                step=0.1,
                                value=4.5,
                                info="Higher values make the image more closely match your prompt.",
                            )
                            num_inference_steps = gr.Slider(
                                label="Number of inference steps",
                                minimum=1,
                                maximum=50,
                                step=1,
                                value=28,
                                info="More steps generally mean higher quality but slower generation.",
                            )
        
        with gr.Column(scale=3):
            with gr.Blocks():
                run_button = gr.Button("Generate", variant="primary", elem_id="generate-button")
            result = gr.Gallery(
                label="Generated Images",
                columns=1,
                height='768px',
                preview=True,
                show_label=True,
            )
            with gr.Accordion(label="Generation Parameters", open=False):
                gr_metadata = gr.JSON(
                    label="Image Metadata",
                    show_label=True,
                )
            gr.Examples(
                examples=examples,
                inputs=prompt,
                outputs=[result, gr_metadata],
                fn=lambda *args, **kwargs: generate(*args, use_upscaler=True, **kwargs),
                cache_examples=CACHE_EXAMPLES,
            )

    use_upscaler.change(
        fn=lambda x: [gr.update(visible=x), gr.update(visible=x)],
        inputs=use_upscaler,
        outputs=[upscaler_strength, upscale_by],
        queue=False,
        api_name=False,
    )
    aspect_ratio_selector.change(
        fn=lambda x: gr.update(visible=x == "Custom"),
        inputs=aspect_ratio_selector,
        outputs=custom_resolution,
        queue=False,
        api_name=False,
    )

    # Combine all triggers including keyboard shortcuts
    gr.on(
        triggers=[
            prompt.submit,
            negative_prompt.submit,
            run_button.click,
        ],
        fn=utils.randomize_seed_fn,
        inputs=[seed, randomize_seed],
        outputs=seed,
        queue=False,
        api_name=False,
    ).then(
        fn=lambda: gr.update(interactive=False, value="Generating..."),
        outputs=run_button,
    ).then(
        fn=generate,
        inputs=[
            prompt,
            negative_prompt,
            seed,
            custom_width,
            custom_height,
            guidance_scale,
            num_inference_steps,
            sampler,
            model_name,
            aspect_ratio_selector,
            use_upscaler,
            upscaler_strength,
            upscale_by,
            add_quality_tags,
        ],
        outputs=[result, gr_metadata],
    ).then(
        fn=lambda: gr.update(interactive=True, value="Generate"),
        outputs=run_button,
    )

if __name__ == "__main__":
    demo.queue(max_size=20).launch(debug=IS_COLAB, share=IS_COLAB)