app.py

import spaces
import gradio as gr
import torch
from diffusers import ZImagePipeline
import os
from pathlib import Path

# Global variable to store the pipeline
pipe = None

def load_model():
    """
    Load the Z-Image Turbo model before inference.
    This ensures the model is downloaded and ready before any generation requests.
    """
    global pipe
    
    if pipe is not None:
        return pipe
    
    print("Loading Z-Image Turbo model...")
    print("This may take a few minutes on first run while the model downloads...")
    
    try:
        # Load the pipeline with optimal settings
        pipe = ZImagePipeline.from_pretrained(
            "Tongyi-MAI/Z-Image-Turbo",
            torch_dtype=torch.bfloat16,
            low_cpu_mem_usage=False,
        )
        
        # Move to GPU if available
        device = "cuda" if torch.cuda.is_available() else "cpu"
        pipe.to(device)
        print(f"Model loaded on {device}")
        
        # Optional: Enable Flash Attention for better efficiency
        try:
            pipe.transformer.set_attention_backend("flash")
            print("Flash Attention enabled")
        except Exception as e:
            print(f"Flash Attention not available: {e}")
            print("Using default attention backend")
        
        print("Model loaded successfully!")
        return pipe
        
    except Exception as e:
        print(f"Error loading model: {e}")
        raise

# Pre-load the model when the app starts
print("Initializing model on startup...")
try:
    load_model()
    print("Model initialization complete!")
except Exception as e:
    print(f"Warning: Could not pre-load model: {e}")
    print("Model will be loaded on first generation request")

@spaces.GPU()
def generate_image(
    prompt,
    progress=gr.Progress(track_tqdm=True)
):
    """
    Generate an image using Z-Image Turbo model.
    
    Args:
        prompt: Text description of the desired image
    
    Returns:
        Generated PIL Image
    """
    global pipe
    
    # Ensure model is loaded
    if pipe is None:
        progress(0, desc="Loading model...")
        load_model()
    
    if not prompt.strip():
        raise gr.Error("Please enter a prompt to generate an image.")
    
    # Determine device
    device = "cuda" if torch.cuda.is_available() else "cpu"
    
    # Set random seed for reproducibility
    generator = torch.Generator(device).manual_seed(42)
    
    # Generate the image with optimal settings
    progress(0.1, desc="Generating image...")
    
    try:
        result = pipe(
            prompt=prompt,
            negative_prompt=None,
            height=1024,
            width=1024,
            num_inference_steps=9,
            guidance_scale=0.0,
            generator=generator,
        )
        
        image = result.images[0]
        progress(1.0, desc="Complete!")
        
        return image
    
    except Exception as e:
        raise gr.Error(f"Generation failed: {str(e)}")

# Create a custom theme based on Soft theme with Apple-inspired colors
custom_theme = gr.themes.Soft(
    primary_hue=gr.themes.colors.blue,
    secondary_hue=gr.themes.colors.slate,
    neutral_hue=gr.themes.colors.gray,
    spacing_size=gr.themes.sizes.spacing_lg,
    radius_size=gr.themes.sizes.radius_md,
    text_size=gr.themes.sizes.text_lg,
    font=[gr.themes.GoogleFont("Inter"), "ui-sans-serif", "system-ui", "sans-serif"],
    font_mono=[gr.themes.GoogleFont("IBM Plex Mono"), "ui-monospace", "Consolas", "monospace"],
).set(
    # Button styling
    button_primary_background_fill="#0071e3",
    button_primary_background_fill_hover="#0077ed",
    button_primary_text_color="white",
    button_large_padding="16px 40px",
    button_large_radius="12px",
    button_shadow="0 2px 8px rgba(0, 113, 227, 0.2)",
    button_shadow_hover="0 4px 12px rgba(0, 113, 227, 0.3)",
    
    # Input styling
    input_background_fill="#fbfbfd",
    input_background_fill_focus="white",
    input_border_color="#d2d2d7",
    input_border_color_focus="#0071e3",
    input_radius="12px",
    input_padding="16px",
    input_shadow_focus="0 0 0 4px rgba(0, 113, 227, 0.1)",
    
    # Container styling
    block_background_fill="white",
    block_border_width="1px",
    block_border_color="#e5e5e7",
    block_radius="12px",
    block_padding="24px",
    
    # Body styling
    body_background_fill="#f5f5f7",
    body_text_color="#1d1d1f",
    
    # Link styling
    link_text_color="#0071e3",
    link_text_color_hover="#0077ed",
)

# Minimal additional CSS for layout refinements
minimal_css = """
.gradio-container {
    max-width: 900px !important;
    margin: 0 auto !important;
}

.main-header {
    text-align: center;
    margin-bottom: 2rem;
    padding-bottom: 2rem;
    border-bottom: 1px solid #e5e5e7;
}

.main-header h1 {
    font-size: 3rem !important;
    font-weight: 600 !important;
    color: #1d1d1f !important;
    margin: 0 0 0.5rem 0 !important;
    letter-spacing: -1px;
}

.main-header .subtitle {
    font-size: 1.25rem !important;
    color: #86868b !important;
    margin: 0.5rem 0 !important;
}

.attribution {
    margin-top: 1rem;
    font-size: 0.875rem;
    color: #86868b;
}

.attribution a {
    color: #0071e3 !important;
    text-decoration: none;
    font-weight: 500;
}

.footer-info {
    text-align: center;
    padding: 2rem 1rem;
    color: #86868b;
    font-size: 0.875rem;
    margin-top: 2rem;
    border-top: 1px solid #e5e5e7;
}

.footer-info p {
    margin: 0.25rem 0;
}

@media (max-width: 768px) {
    .main-header h1 {
        font-size: 2rem !important;
    }
    
    .main-header .subtitle {
        font-size: 1rem !important;
    }
}
"""

# Create the Gradio interface
with gr.Blocks(
    title="Z-Image Turbo",
    theme=custom_theme,
    css=minimal_css,
    fill_height=False
) as demo:
    
    # Header
    with gr.Column(elem_classes="main-header"):
        gr.Markdown(
            """
            # Z-Image Turbo
            ### Create stunning images from text
            """,
            elem_classes="main-header"
        )
        gr.HTML("""
            <div class="attribution">
                Built with <a href="https://huggingface.co/spaces/akhaliq/anycoder" target="_blank">anycoder</a>
            </div>
        """)
    
    # Prompt input
    prompt = gr.Textbox(
        placeholder="Describe the image you want to create...",
        lines=3,
        max_lines=6,
        label="Prompt",
        show_label=False,
        container=True
    )
    
    # Generate button
    generate_btn = gr.Button(
        "Generate",
        variant="primary",
        size="lg",
        scale=1
    )
    
    # Output image
    output_image = gr.Image(
        type="pil",
        label="Generated Image",
        show_label=False,
        show_download_button=True,
        show_share_button=False,
        container=True
    )
    
    # Footer
    gr.HTML("""
        <div class="footer-info">
            <p>Powered by Z-Image Turbo from Tongyi-MAI</p>
            <p>Optimized for fast, high-quality image generation</p>
        </div>
    """)
    
    # Event handlers
    generate_btn.click(
        fn=generate_image,
        inputs=prompt,
        outputs=output_image,
        api_name="generate"
    )
    
    # Also allow generation on Enter key
    prompt.submit(
        fn=generate_image,
        inputs=prompt,
        outputs=output_image
    )

# Launch the app
if __name__ == "__main__":
    demo.launch(
        share=False,
        show_error=True
    )