Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,249 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
import numpy as np
|
| 3 |
+
import random
|
| 4 |
+
import torch
|
| 5 |
+
import spaces
|
| 6 |
+
from PIL import Image
|
| 7 |
+
from huggingface_hub import hf_hub_download
|
| 8 |
+
from safetensors.torch import load_file
|
| 9 |
+
from tqdm import tqdm
|
| 10 |
+
import gc
|
| 11 |
+
|
| 12 |
+
from qwenimage.pipeline_qwen_image_edit import QwenImageEditPipeline
|
| 13 |
+
from qwenimage.transformer_qwenimage import QwenImageTransformer2DModel
|
| 14 |
+
from qwenimage.qwen_fa3_processor import QwenDoubleStreamAttnProcessorFA3
|
| 15 |
+
|
| 16 |
+
|
| 17 |
+
LORA_CONFIG = {
|
| 18 |
+
"None": {
|
| 19 |
+
"repo_id": None,
|
| 20 |
+
"filename": None,
|
| 21 |
+
"type": "edit",
|
| 22 |
+
"method": "none",
|
| 23 |
+
"prompt_template": "{prompt}",
|
| 24 |
+
"description": "Use the base Qwen-Image-Edit model without any LoRA.",
|
| 25 |
+
},
|
| 26 |
+
"InStyle (Style Transfer)": {
|
| 27 |
+
"repo_id": "peteromallet/Qwen-Image-Edit-InStyle",
|
| 28 |
+
"filename": "InStyle-0.5.safetensors",
|
| 29 |
+
"type": "style",
|
| 30 |
+
"method": "manual_fuse",
|
| 31 |
+
"prompt_template": "Make an image in this style of {prompt}",
|
| 32 |
+
"description": "Transfers the style from a reference image to a new image described by the prompt.",
|
| 33 |
+
},
|
| 34 |
+
"InScene (In-Scene Editing)": {
|
| 35 |
+
"repo_id": "flymy-ai/qwen-image-edit-inscene-lora",
|
| 36 |
+
"filename": "flymy_qwen_image_edit_inscene_lora.safetensors",
|
| 37 |
+
"type": "edit",
|
| 38 |
+
"method": "standard",
|
| 39 |
+
"prompt_template": "{prompt}",
|
| 40 |
+
"description": "Improves in-scene editing, object positioning, and camera perspective changes.",
|
| 41 |
+
},
|
| 42 |
+
"Face Segmentation": {
|
| 43 |
+
"repo_id": "TsienDragon/qwen-image-edit-lora-face-segmentation",
|
| 44 |
+
"filename": "pytorch_lora_weights.safetensors",
|
| 45 |
+
"type": "edit",
|
| 46 |
+
"method": "standard",
|
| 47 |
+
"prompt_template": "change the face to face segmentation mask",
|
| 48 |
+
"description": "Transforms a facial image into a precise segmentation mask.",
|
| 49 |
+
},
|
| 50 |
+
"Object Remover": {
|
| 51 |
+
"repo_id": "valiantcat/Qwen-Image-Edit-Remover-General-LoRA",
|
| 52 |
+
"filename": "qwen-edit-remover.safetensors",
|
| 53 |
+
"type": "edit",
|
| 54 |
+
"method": "standard",
|
| 55 |
+
"prompt_template": "Remove {prompt}",
|
| 56 |
+
"description": "Removes objects from an image while maintaining background consistency.",
|
| 57 |
+
},
|
| 58 |
+
}
|
| 59 |
+
|
| 60 |
+
print("Initializing model...")
|
| 61 |
+
dtype = torch.bfloat16
|
| 62 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
| 63 |
+
|
| 64 |
+
pipe = QwenImageEditPipeline.from_pretrained(
|
| 65 |
+
"Qwen/Qwen-Image-Edit",
|
| 66 |
+
torch_dtype=dtype
|
| 67 |
+
).to(device)
|
| 68 |
+
|
| 69 |
+
pipe.transformer.__class__ = QwenImageTransformer2DModel
|
| 70 |
+
pipe.transformer.set_attn_processor(QwenDoubleStreamAttnProcessorFA3())
|
| 71 |
+
|
| 72 |
+
original_transformer_state_dict = pipe.transformer.state_dict()
|
| 73 |
+
print("Base model loaded and ready.")
|
| 74 |
+
|
| 75 |
+
def fuse_lora_manual(transformer, lora_state_dict, alpha=1.0):
|
| 76 |
+
key_mapping = {}
|
| 77 |
+
for key in lora_state_dict.keys():
|
| 78 |
+
base_key = key.replace('diffusion_model.', '').rsplit('.lora_', 1)[0]
|
| 79 |
+
if base_key not in key_mapping:
|
| 80 |
+
key_mapping[base_key] = {}
|
| 81 |
+
if 'lora_A' in key:
|
| 82 |
+
key_mapping[base_key]['down'] = lora_state_dict[key]
|
| 83 |
+
elif 'lora_B' in key:
|
| 84 |
+
key_mapping[base_key]['up'] = lora_state_dict[key]
|
| 85 |
+
|
| 86 |
+
for name, module in tqdm(transformer.named_modules(), desc="Fusing layers"):
|
| 87 |
+
if name in key_mapping and isinstance(module, torch.nn.Linear):
|
| 88 |
+
lora_weights = key_mapping[name]
|
| 89 |
+
if 'down' in lora_weights and 'up' in lora_weights:
|
| 90 |
+
device = module.weight.device
|
| 91 |
+
dtype = module.weight.dtype
|
| 92 |
+
lora_down = lora_weights['down'].to(device, dtype=dtype)
|
| 93 |
+
lora_up = lora_weights['up'].to(device, dtype=dtype)
|
| 94 |
+
merged_delta = lora_up @ lora_down
|
| 95 |
+
module.weight.data += alpha * merged_delta
|
| 96 |
+
return transformer
|
| 97 |
+
|
| 98 |
+
def load_and_fuse_lora(lora_name):
|
| 99 |
+
"""Carrega uma LoRA, funde-a ao modelo e retorna o pipeline modificado."""
|
| 100 |
+
config = LORA_CONFIG[lora_name]
|
| 101 |
+
|
| 102 |
+
print("Resetting transformer to original state...")
|
| 103 |
+
pipe.transformer.load_state_dict(original_transformer_state_dict)
|
| 104 |
+
|
| 105 |
+
if config["method"] == "none":
|
| 106 |
+
print("No LoRA selected. Using base model.")
|
| 107 |
+
return
|
| 108 |
+
|
| 109 |
+
print(f"Loading LoRA: {lora_name}")
|
| 110 |
+
lora_path = hf_hub_download(repo_id=config["repo_id"], filename=config["filename"])
|
| 111 |
+
|
| 112 |
+
if config["method"] == "standard":
|
| 113 |
+
print("Using standard loading method...")
|
| 114 |
+
pipe.load_lora_weights(lora_path)
|
| 115 |
+
print("Fusing LoRA into the model...")
|
| 116 |
+
pipe.fuse_lora()
|
| 117 |
+
elif config["method"] == "manual_fuse":
|
| 118 |
+
print("Using manual fusion method...")
|
| 119 |
+
lora_state_dict = load_file(lora_path)
|
| 120 |
+
pipe.transformer = fuse_lora_manual(pipe.transformer, lora_state_dict)
|
| 121 |
+
|
| 122 |
+
gc.collect()
|
| 123 |
+
torch.cuda.empty_cache()
|
| 124 |
+
print(f"LoRA '{lora_name}' is now active.")
|
| 125 |
+
|
| 126 |
+
@spaces.GPU(duration=60)
|
| 127 |
+
def infer(
|
| 128 |
+
lora_name,
|
| 129 |
+
input_image,
|
| 130 |
+
style_image,
|
| 131 |
+
prompt,
|
| 132 |
+
seed,
|
| 133 |
+
randomize_seed,
|
| 134 |
+
true_guidance_scale,
|
| 135 |
+
num_inference_steps,
|
| 136 |
+
progress=gr.Progress(track_tqdm=True),
|
| 137 |
+
):
|
| 138 |
+
if not lora_name:
|
| 139 |
+
raise gr.Error("Please select a LoRA model.")
|
| 140 |
+
|
| 141 |
+
config = LORA_CONFIG[lora_name]
|
| 142 |
+
|
| 143 |
+
if config["type"] == "style":
|
| 144 |
+
if style_image is None:
|
| 145 |
+
raise gr.Error("Style Transfer LoRA requires a Style Reference Image.")
|
| 146 |
+
image_for_pipeline = style_image
|
| 147 |
+
else: # 'edit'
|
| 148 |
+
if input_image is None:
|
| 149 |
+
raise gr.Error("This LoRA requires an Input Image.")
|
| 150 |
+
image_for_pipeline = input_image
|
| 151 |
+
|
| 152 |
+
if not prompt and config["prompt_template"] != "change the face to face segmentation mask":
|
| 153 |
+
raise gr.Error("A text prompt is required for this LoRA.")
|
| 154 |
+
|
| 155 |
+
load_and_fuse_lora(lora_name)
|
| 156 |
+
|
| 157 |
+
final_prompt = config["prompt_template"].format(prompt=prompt)
|
| 158 |
+
|
| 159 |
+
if randomize_seed:
|
| 160 |
+
seed = random.randint(0, np.iinfo(np.int32).max)
|
| 161 |
+
generator = torch.Generator(device=device).manual_seed(int(seed))
|
| 162 |
+
|
| 163 |
+
print("--- Running Inference ---")
|
| 164 |
+
print(f"LoRA: {lora_name}")
|
| 165 |
+
print(f"Prompt: {final_prompt}")
|
| 166 |
+
print(f"Seed: {seed}, Steps: {num_inference_steps}, CFG: {true_guidance_scale}")
|
| 167 |
+
|
| 168 |
+
with torch.inference_mode():
|
| 169 |
+
result_image = pipe(
|
| 170 |
+
image=image_for_pipeline,
|
| 171 |
+
prompt=final_prompt,
|
| 172 |
+
negative_prompt=" ",
|
| 173 |
+
num_inference_steps=int(num_inference_steps),
|
| 174 |
+
generator=generator,
|
| 175 |
+
true_cfg_scale=true_guidance_scale,
|
| 176 |
+
).images[0]
|
| 177 |
+
|
| 178 |
+
pipe.unfuse_lora()
|
| 179 |
+
gc.collect()
|
| 180 |
+
torch.cuda.empty_cache()
|
| 181 |
+
|
| 182 |
+
return result_image, seed
|
| 183 |
+
|
| 184 |
+
def on_lora_change(lora_name):
|
| 185 |
+
config = LORA_CONFIG[lora_name]
|
| 186 |
+
is_style_lora = config["type"] == "style"
|
| 187 |
+
return {
|
| 188 |
+
lora_description: gr.Markdown(visible=True, value=f"**Description:** {config['description']}"),
|
| 189 |
+
input_image_box: gr.Image(visible=not is_style_lora),
|
| 190 |
+
style_image_box: gr.Image(visible=is_style_lora),
|
| 191 |
+
prompt_box: gr.Textbox(visible=(config["prompt_template"] != "change the face to face segmentation mask"))
|
| 192 |
+
}
|
| 193 |
+
|
| 194 |
+
with gr.Blocks(css="#col-container { margin: 0 auto; max-width: 1024px; }") as demo:
|
| 195 |
+
with gr.Column(elem_id="col-container"):
|
| 196 |
+
gr.HTML('<img src="https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-Image/qwen_image_edit_logo.png" alt="Qwen-Image Logo" style="width: 400px; margin: 0 auto; display: block;">')
|
| 197 |
+
gr.Markdown("<h2 style='text-align: center;'>Qwen-Image-Edit Multi-LoRA Playground</h2>")
|
| 198 |
+
|
| 199 |
+
with gr.Row():
|
| 200 |
+
with gr.Column(scale=1):
|
| 201 |
+
lora_selector = gr.Dropdown(
|
| 202 |
+
label="Select LoRA Model",
|
| 203 |
+
choices=list(LORA_CONFIG.keys()),
|
| 204 |
+
value="InStyle (Style Transfer)"
|
| 205 |
+
)
|
| 206 |
+
lora_description = gr.Markdown(visible=False)
|
| 207 |
+
|
| 208 |
+
input_image_box = gr.Image(label="Input Image", type="pil", visible=False)
|
| 209 |
+
style_image_box = gr.Image(label="Style Reference Image", type="pil", visible=True)
|
| 210 |
+
|
| 211 |
+
prompt_box = gr.Textbox(label="Prompt", placeholder="Describe the content or object to remove...")
|
| 212 |
+
|
| 213 |
+
run_button = gr.Button("Generate!", variant="primary")
|
| 214 |
+
|
| 215 |
+
with gr.Column(scale=1):
|
| 216 |
+
result_image = gr.Image(label="Result", type="pil")
|
| 217 |
+
used_seed = gr.Number(label="Used Seed", interactive=False)
|
| 218 |
+
|
| 219 |
+
with gr.Accordion("Advanced Settings", open=False):
|
| 220 |
+
seed_slider = gr.Slider(label="Seed", minimum=0, maximum=np.iinfo(np.int32).max, step=1, value=42)
|
| 221 |
+
randomize_seed_checkbox = gr.Checkbox(label="Randomize seed", value=True)
|
| 222 |
+
cfg_slider = gr.Slider(label="Guidance Scale (CFG)", minimum=1.0, maximum=10.0, step=0.1, value=4.0)
|
| 223 |
+
steps_slider = gr.Slider(label="Inference Steps", minimum=10, maximum=50, step=1, value=25)
|
| 224 |
+
|
| 225 |
+
lora_selector.change(
|
| 226 |
+
fn=on_lora_change,
|
| 227 |
+
inputs=lora_selector,
|
| 228 |
+
outputs=[lora_description, input_image_box, style_image_box, prompt_box]
|
| 229 |
+
).then(
|
| 230 |
+
None,
|
| 231 |
+
lora_selector,
|
| 232 |
+
[lora_description, input_image_box, style_image_box, prompt_box],
|
| 233 |
+
_js="() => { document.querySelector('#lora_selector select').dispatchEvent(new Event('change')) }"
|
| 234 |
+
)
|
| 235 |
+
|
| 236 |
+
run_button.click(
|
| 237 |
+
fn=infer,
|
| 238 |
+
inputs=[
|
| 239 |
+
lora_selector,
|
| 240 |
+
input_image_box, style_image_box,
|
| 241 |
+
prompt_box,
|
| 242 |
+
seed_slider, randomize_seed_checkbox,
|
| 243 |
+
cfg_slider, steps_slider
|
| 244 |
+
],
|
| 245 |
+
outputs=[result_image, used_seed]
|
| 246 |
+
)
|
| 247 |
+
|
| 248 |
+
if __name__ == "__main__":
|
| 249 |
+
demo.launch()
|