app.py

import gradio as gr
import numpy as np
import random
import spaces
import torch
from diffusers import QwenImagePipeline

dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"

pipe = QwenImagePipeline.from_pretrained("Qwen/Qwen-Image", torch_dtype=dtype).to(device)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1536

@spaces.GPU()
def infer(prompt, negative_prompt, seed=42, randomize_seed=False, width=1024, height=1024, num_inference_steps=4, true_cfg_scale=4.0, distilled_cfg_scale=1.0, progress=gr.Progress(track_tqdm=True)):
    """
    Generates an image based on a user's prompt using the Qwen-Image pipeline.

    This function takes textual prompts and various generation parameters,
    handles seed randomization, and runs the diffusion model to produce an image.

    Args:
        prompt (str): The positive text prompt to guide image generation.
        negative_prompt (str): The negative text prompt to guide the model
            on what to avoid in the generated image.
        seed (int, optional): The seed for the random number generator to ensure
            reproducible results. Defaults to 42.
        randomize_seed (bool, optional): If True, a random seed is generated,
            overriding the `seed` parameter. Defaults to False.
        width (int, optional): The width of the generated image in pixels.
            Defaults to 1024.
        height (int, optional): The height of the generated image in pixels.
            Defaults to 1024.
        num_inference_steps (int, optional): The number of denoising steps.
            More steps can lead to higher quality but take longer. Defaults to 4.
        true_cfg_scale (float, optional): The Classifier-Free Guidance scale.
            Controls how strictly the model follows the prompt. Defaults to 4.0.
        progress (gr.Progress, optional): A Gradio Progress object to track
            the inference progress in the UI.

    Returns:
        tuple: A tuple containing:
            - PIL.Image.Image: The generated image.
            - int: The seed used for the generation, which is useful for
              reproducibility, especially when `randomize_seed` is True.
    """
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    
    generator = torch.Generator().manual_seed(seed)
    
    image = pipe(
            prompt=prompt,
            negative_prompt=negative_prompt,
            width=width,
            height=height,
            num_inference_steps=num_inference_steps, 
            generator=generator,
            true_cfg_scale=true_cfg_scale,
            guidance_scale=distilled_cfg_scale
    ).images[0] 
    
    return image, seed
 
examples = [
    ["a tiny dragon hatching from a crystal egg on Mars"],
    ["a red panda holding a sign that says 'I love bamboo'"],
    ["a photo of a capybara riding a tricycle in Paris. It is wearing a beret and a striped shirt."],
    ["an anime illustration of a delicious ramen bowl"],
    ["A logo for a bookstore called 'The Whispering Page'. The logo should feature an open book with a tree growing out of it."],
]

css="""
#col-container {
    margin: 0 auto;
    max-width: 580px;
}
"""

# Build the Gradio UI.
with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        # Title and description for the demo.
        gr.Markdown(f"""# Qwen-Image Text-to-Image
Gradio demo for [Qwen-Image](https://huggingface.co/Qwen/Qwen-Image), a powerful text-to-image model from the Qwen team at Alibaba.
        """)
        
        with gr.Row():
            # Main prompt input.
            prompt = gr.Text(
                label="Prompt",
                show_label=False,
                max_lines=1,
                placeholder="Enter your prompt",
                container=False,
            )
            # The "Run" button.
            run_button = gr.Button("Run", scale=0)
        
        
        result = gr.Image(label="Result", show_label=False)
        
        negative_prompt = gr.Text(
            label="Negative Prompt",
            max_lines=1,
            placeholder="Enter a negative prompt",
            value="text, watermark, copyright, blurry, low resolution",
        )
        # Accordion for advanced settings.
        with gr.Accordion("Advanced Settings", open=False):
            
            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=42,
            )
            
            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
            
            with gr.Row():
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
            
            with gr.Row():
                num_inference_steps = gr.Slider(
                    label="Inference Steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=28,
                )
                true_cfg_scale = gr.Slider(
                    label="CFG Scale",
                    info="Controls how much the model follows the prompt. Higher values mean stricter adherence.",
                    minimum=1.0,
                    maximum=10.0,
                    step=0.1,
                    value=4.0
                )
                distilled_cfg_scale = gr.Slider(
                    label="Distilled Guidance",
                    minimum=0.0,
                    maximum=20.0,
                    step=0.1,
                    value=1.0
                )
        
        gr.Examples(
            examples=examples,
            fn=infer,
            inputs=[prompt, negative_prompt],
            outputs=[result, seed],
            cache_examples=True,
            cache_mode='lazy'
        )

    gr.on(
        triggers=[run_button.click, prompt.submit, negative_prompt.submit],
        fn=infer,
        inputs=[prompt, negative_prompt, seed, randomize_seed, width, height, num_inference_steps, true_cfg_scale, distilled_cfg_scale],
        outputs=[result, seed]
    )

demo.launch(mcp_server=True)