app.py

import gradio as gr
import numpy as np
import random
import spaces
import torch
import time
import os
from diffusers import DiffusionPipeline
from custom_pipeline import FLUXPipelineWithIntermediateOutputs
from transformers import pipeline

# Translation model loading with device specification
translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en", device="cpu")

# Constants
MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048
DEFAULT_WIDTH = 1024
DEFAULT_HEIGHT = 1024
DEFAULT_INFERENCE_STEPS = 1
GPU_DURATION = 15  # Reduced from 25 to stay within quota

# Device and model setup with memory optimization
def setup_model():
    dtype = torch.float16
    pipe = FLUXPipelineWithIntermediateOutputs.from_pretrained(
        "black-forest-labs/FLUX.1-schnell", 
        torch_dtype=dtype,
        device_map="auto"  # Enable model parallelism
    )
    return pipe

pipe = setup_model()

# Menu labels dictionary
english_labels = {
    "Generated Image": "Generated Image",
    "Prompt": "Prompt",
    "Enhance Image": "Enhance Image",
    "Advanced Options": "Advanced Options",
    "Seed": "Seed",
    "Randomize Seed": "Randomize Seed",
    "Width": "Width",
    "Height": "Height",
    "Inference Steps": "Inference Steps",
    "Inspiration Gallery": "Inspiration Gallery"
}

def translate_if_korean(text):
    """Safely translate Korean text to English."""
    try:
        if any('\u3131' <= char <= '\u3163' or '\uac00' <= char <= '\ud7a3' for char in text):
            return translator(text)[0]['translation_text']
        return text
    except Exception as e:
        print(f"Translation error: {e}")
        return text

# Modified inference function with error handling and memory management
@spaces.GPU(duration=GPU_DURATION)
def generate_image(prompt, seed=None, width=DEFAULT_WIDTH, height=DEFAULT_HEIGHT, 
                  randomize_seed=True, num_inference_steps=DEFAULT_INFERENCE_STEPS):
    try:
        # Input validation
        if not isinstance(seed, (int, type(None))):
            seed = None
            randomize_seed = True
            
        prompt = translate_if_korean(prompt)
        
        if seed is None or randomize_seed:
            seed = random.randint(0, MAX_SEED)
        
        # Ensure valid dimensions
        width = min(max(256, width), MAX_IMAGE_SIZE)
        height = min(max(256, height), MAX_IMAGE_SIZE)
        
        generator = torch.Generator().manual_seed(seed)
        
        start_time = time.time()
        
        with torch.cuda.amp.autocast():  # Enable automatic mixed precision
            for img in pipe.generate_images(
                prompt=prompt,
                guidance_scale=0,
                num_inference_steps=num_inference_steps,
                width=width,
                height=height,
                generator=generator
            ):
                latency = f"Processing Time: {(time.time()-start_time):.2f} seconds"
                
                # Clear CUDA cache after generation
                if torch.cuda.is_available():
                    torch.cuda.empty_cache()
                    
                yield img, seed, latency
                
    except Exception as e:
        print(f"Error in generate_image: {e}")
        # Return a blank image or error message
        yield None, seed, f"Error: {str(e)}"

# Example generator with error handling
def generate_example_image(prompt):
    try:
        return next(generate_image(prompt, randomize_seed=True))
    except Exception as e:
        print(f"Error in example generation: {e}")
        return None, None, f"Error: {str(e)}"

# Example prompts
examples = [
    "비너 슈니첼의 애니메이션 일러스트레이션",
    "A steampunk owl wearing Victorian-era clothing and reading a mechanical book",
    "A floating island made of books with waterfalls of knowledge cascading down",
    "A bioluminescent forest where mushrooms glow like neon signs in a cyberpunk city",
    "An ancient temple being reclaimed by nature, with robots performing archaeology",
    "A cosmic coffee shop where baristas are constellations serving drinks made of stardust"
]

css = """
footer {
    visibility: hidden;
}
"""

# --- Gradio UI with improved error handling ---
with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
    with gr.Column(elem_id="app-container"):
        with gr.Row():
            with gr.Column(scale=3):
                result = gr.Image(label=english_labels["Generated Image"], 
                                show_label=False, 
                                interactive=False)
            with gr.Column(scale=1):
                prompt = gr.Text(
                    label=english_labels["Prompt"],
                    placeholder="Describe the image you want to generate...",
                    lines=3,
                    show_label=False,
                    container=False,
                )
                enhanceBtn = gr.Button(f"🚀 {english_labels['Enhance Image']}")

                with gr.Column(english_labels["Advanced Options"]):
                    with gr.Row():
                        latency = gr.Text(show_label=False)
                    with gr.Row():
                        # Modified Number component with proper validation
                        seed = gr.Number(
                            label=english_labels["Seed"], 
                            value=42, 
                            precision=0,
                            minimum=0,
                            maximum=MAX_SEED
                        )
                        randomize_seed = gr.Checkbox(
                            label=english_labels["Randomize Seed"], 
                            value=True
                        )
                    with gr.Row():
                        width = gr.Slider(
                            label=english_labels["Width"], 
                            minimum=256, 
                            maximum=MAX_IMAGE_SIZE, 
                            step=32, 
                            value=DEFAULT_WIDTH
                        )
                        height = gr.Slider(
                            label=english_labels["Height"], 
                            minimum=256, 
                            maximum=MAX_IMAGE_SIZE, 
                            step=32, 
                            value=DEFAULT_HEIGHT
                        )
                        num_inference_steps = gr.Slider(
                            label=english_labels["Inference Steps"], 
                            minimum=1, 
                            maximum=4, 
                            step=1, 
                            value=DEFAULT_INFERENCE_STEPS
                        )

        with gr.Row():
            gr.Markdown(f"### 🌟 {english_labels['Inspiration Gallery']}")
        with gr.Row():
            gr.Examples(
                examples=examples,
                fn=generate_example_image,
                inputs=[prompt],
                outputs=[result, seed],
                cache_examples=False
            )

    # Event handling with improved error handling
    enhanceBtn.click(
        fn=generate_image,
        inputs=[prompt, seed, width, height],
        outputs=[result, seed, latency],
        show_progress="hidden",
        show_api=False,
        queue=False
    )

    # Modified event handler with proper input validation
    def validated_generate(*args):
        try:
            return next(generate_image(*args))
        except Exception as e:
            print(f"Error in validated_generate: {e}")
            return None, args[1], f"Error: {str(e)}"

    gr.on(
        triggers=[prompt.input, width.input, height.input, num_inference_steps.input],
        fn=validated_generate,
        inputs=[prompt, seed, width, height, randomize_seed, num_inference_steps],
        outputs=[result, seed, latency],
        show_progress="hidden",
        show_api=False,
        trigger_mode="always_last",
        queue=False
    )

if __name__ == "__main__":
    demo.launch()