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 (通义千问) 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) # Negative prompt input. negative_prompt = gr.Text( label="Negative Prompt", max_lines=1, placeholder="Enter a negative prompt", value="text, watermark, copyright, blurry, low resolution", ) # Display area for the generated image. result = gr.Image(label="Result", show_label=False) # 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=4, ) 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="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)