Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,137 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import numpy as np
|
| 3 |
+
import random
|
| 4 |
+
import torch
|
| 5 |
+
import spaces
|
| 6 |
+
|
| 7 |
+
from PIL import Image
|
| 8 |
+
from diffusers import QwenImageEditPipeline
|
| 9 |
+
|
| 10 |
+
import os
|
| 11 |
+
|
| 12 |
+
# --- Model Loading ---
|
| 13 |
+
dtype = torch.bfloat16
|
| 14 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 15 |
+
|
| 16 |
+
# Load the model pipeline
|
| 17 |
+
pipe = QwenImageEditPipeline.from_pretrained("Qwen/Qwen-Image-Edit", torch_dtype=dtype).to(device)
|
| 18 |
+
|
| 19 |
+
# --- UI Constants and Helpers ---
|
| 20 |
+
MAX_SEED = np.iinfo(np.int32).max
|
| 21 |
+
|
| 22 |
+
# --- Main Inference Function (with hardcoded negative prompt) ---
|
| 23 |
+
@spaces.GPU(duration=120)
|
| 24 |
+
def infer(
|
| 25 |
+
image,
|
| 26 |
+
prompt,
|
| 27 |
+
seed=42,
|
| 28 |
+
randomize_seed=False,
|
| 29 |
+
guidance_scale=4.0,
|
| 30 |
+
num_inference_steps=50,
|
| 31 |
+
progress=gr.Progress(track_tqdm=True),
|
| 32 |
+
):
|
| 33 |
+
"""
|
| 34 |
+
Generates an image using the local Qwen-Image diffusers pipeline.
|
| 35 |
+
"""
|
| 36 |
+
# Hardcode the negative prompt as requested
|
| 37 |
+
negative_prompt = "text, watermark, copyright, blurry, low resolution"
|
| 38 |
+
|
| 39 |
+
if randomize_seed:
|
| 40 |
+
seed = random.randint(0, MAX_SEED)
|
| 41 |
+
|
| 42 |
+
# Set up the generator for reproducibility
|
| 43 |
+
generator = torch.Generator(device=device).manual_seed(seed)
|
| 44 |
+
|
| 45 |
+
print(f"Calling pipeline with prompt: '{prompt}'")
|
| 46 |
+
print(f"Negative Prompt: '{negative_prompt}'")
|
| 47 |
+
print(f"Seed: {seed}, Steps: {num_inference_steps}, Guidance: {guidance_scale}")
|
| 48 |
+
|
| 49 |
+
# Generate the image
|
| 50 |
+
image = pipe(
|
| 51 |
+
image,
|
| 52 |
+
prompt=prompt,
|
| 53 |
+
negative_prompt=negative_prompt,
|
| 54 |
+
num_inference_steps=num_inference_steps,
|
| 55 |
+
generator=generator,
|
| 56 |
+
true_cfg_scale=guidance_scale,
|
| 57 |
+
guidance_scale=1.0 # Use a fixed default for distilled guidance
|
| 58 |
+
).images[0]
|
| 59 |
+
|
| 60 |
+
return image, seed
|
| 61 |
+
|
| 62 |
+
# --- Examples and UI Layout ---
|
| 63 |
+
examples = []
|
| 64 |
+
|
| 65 |
+
css = """
|
| 66 |
+
#col-container {
|
| 67 |
+
margin: 0 auto;
|
| 68 |
+
max-width: 1024px;
|
| 69 |
+
}
|
| 70 |
+
"""
|
| 71 |
+
|
| 72 |
+
with gr.Blocks(css=css) as demo:
|
| 73 |
+
with gr.Column(elem_id="col-container"):
|
| 74 |
+
gr.Markdown('<img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Image/qwen_image_logo.png" alt="Qwen-Image Logo" width="400" style="display: block; margin: 0 auto;">')
|
| 75 |
+
gr.Markdown("[Learn more](https://github.com/QwenLM/Qwen-Image) about the Qwen-Image series. Try on [Qwen Chat](https://chat.qwen.ai/), or [download model](https://huggingface.co/Qwen/Qwen-Image-Edit) to run locally with ComfyUI or diffusers.")
|
| 76 |
+
with gr.Row():
|
| 77 |
+
with gr.Column():
|
| 78 |
+
input_image = gr.Image(label="Input Image", show_label=False, type="pil")
|
| 79 |
+
prompt = gr.Text(
|
| 80 |
+
label="Prompt",
|
| 81 |
+
show_label=False,
|
| 82 |
+
placeholder="describe the edit instruction",
|
| 83 |
+
container=False,
|
| 84 |
+
)
|
| 85 |
+
run_button = gr.Button("Run", scale=0, variant="primary")
|
| 86 |
+
|
| 87 |
+
result = gr.Image(label="Result", show_label=False, type="pil")
|
| 88 |
+
|
| 89 |
+
with gr.Accordion("Advanced Settings", open=False):
|
| 90 |
+
# Negative prompt UI element is removed here
|
| 91 |
+
|
| 92 |
+
seed = gr.Slider(
|
| 93 |
+
label="Seed",
|
| 94 |
+
minimum=0,
|
| 95 |
+
maximum=MAX_SEED,
|
| 96 |
+
step=1,
|
| 97 |
+
value=0,
|
| 98 |
+
)
|
| 99 |
+
|
| 100 |
+
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
| 101 |
+
|
| 102 |
+
with gr.Row():
|
| 103 |
+
guidance_scale = gr.Slider(
|
| 104 |
+
label="Guidance scale",
|
| 105 |
+
minimum=0.0,
|
| 106 |
+
maximum=10.0,
|
| 107 |
+
step=0.1,
|
| 108 |
+
value=4.0,
|
| 109 |
+
)
|
| 110 |
+
|
| 111 |
+
num_inference_steps = gr.Slider(
|
| 112 |
+
label="Number of inference steps",
|
| 113 |
+
minimum=1,
|
| 114 |
+
maximum=50,
|
| 115 |
+
step=1,
|
| 116 |
+
value=30,
|
| 117 |
+
)
|
| 118 |
+
|
| 119 |
+
# gr.Examples(examples=examples, inputs=[prompt], outputs=[result, seed], fn=infer, cache_examples=False)
|
| 120 |
+
|
| 121 |
+
gr.on(
|
| 122 |
+
triggers=[run_button.click, prompt.submit],
|
| 123 |
+
fn=infer,
|
| 124 |
+
inputs=[
|
| 125 |
+
input_image,
|
| 126 |
+
prompt,
|
| 127 |
+
# negative_prompt is no longer an input from the UI
|
| 128 |
+
seed,
|
| 129 |
+
randomize_seed,
|
| 130 |
+
guidance_scale,
|
| 131 |
+
num_inference_steps,
|
| 132 |
+
],
|
| 133 |
+
outputs=[result, seed],
|
| 134 |
+
)
|
| 135 |
+
|
| 136 |
+
if __name__ == "__main__":
|
| 137 |
+
demo.launch()
|