app.py

import gradio as gr
import numpy as np
import spaces
import torch
import random
import os

from diffusers import QwenImageEditInpaintPipeline
from PIL import Image

# Set environment variable for parallel loading
os.environ["HF_ENABLE_PARALLEL_LOADING"] = "YES"

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

# Initialize Qwen Image Edit pipeline
pipe = QwenImageEditInpaintPipeline.from_pretrained("Qwen/Qwen-Image-Edit", torch_dtype=torch.bfloat16).to("cuda")

@spaces.GPU
def infer(edit_images, prompt, negative_prompt="", seed=42, randomize_seed=False, strength=1.0, num_inference_steps=35, true_cfg_scale=4.0, progress=gr.Progress(track_tqdm=True)):
    image = edit_images["background"]
    mask = edit_images["layers"][0]
    
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    
    # Generate image using Qwen pipeline
    result_image = pipe(
        prompt=prompt,
        negative_prompt=negative_prompt,
        image=image,
        mask_image=mask,
        strength=strength,
        num_inference_steps=num_inference_steps,
        true_cfg_scale=true_cfg_scale,
        generator=torch.Generator(device="cuda").manual_seed(seed)
    ).images[0]
    
    return result_image, seed
    
examples = [
    "change the hat to red",
    "make the background a beautiful sunset",
    "replace the object with a flower vase",
]

css="""
#col-container {
    margin: 0 auto;
    max-width: 1000px;
}
"""

with gr.Blocks(css=css) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""# Qwen Image Edit Inpainting
Advanced image inpainting using Qwen's Image Edit model  
[[model](https://huggingface.co/Qwen/Qwen-Image-Edit)] [[paper](https://arxiv.org/abs/2412.20710)]
        """)
        with gr.Row():
            with gr.Column():
                edit_image = gr.ImageEditor(
                    label='Upload and draw mask for inpainting',
                    type='pil',
                    sources=["upload", "webcam"],
                    image_mode='RGB',
                    layers=False,
                    brush=gr.Brush(colors=["#FFFFFF"], color_mode="fixed"),
                    height=600
                )
                prompt = gr.Text(
                    label="Prompt",
                    show_label=False,
                    max_lines=1,
                    placeholder="Enter your prompt (e.g., 'change the hat to red')",
                    container=False,
                )
                negative_prompt = gr.Text(
                    label="Negative Prompt",
                    show_label=True,
                    max_lines=1,
                    placeholder="Enter what you don't want (optional)",
                    container=False,
                    value=""
                )
                run_button = gr.Button("Run")
                
            result = gr.Image(label="Result", show_label=False)
        
        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():
                strength = gr.Slider(
                    label="Strength",
                    minimum=0.0,
                    maximum=2.0,
                    step=0.1,
                    value=1.0,
                    info="Controls how much the inpainted region should change"
                )
                
                true_cfg_scale = gr.Slider(
                    label="True CFG Scale",
                    minimum=1.0,
                    maximum=20.0,
                    step=0.5,
                    value=4.0,
                    info="Classifier-free guidance scale"
                )
  
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=10,
                    maximum=100,
                    step=1,
                    value=35,
                )

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

demo.launch()