Spaces:
Running
on
Zero
Running
on
Zero
添加prompt tab
Browse files- app.py +126 -2
- ominicontrol.py +89 -15
app.py
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
import spaces
|
| 3 |
-
from ominicontrol import generate_image
|
| 4 |
import os
|
| 5 |
|
| 6 |
from huggingface_hub import login
|
|
@@ -193,10 +193,134 @@ def infer(
|
|
| 193 |
)
|
| 194 |
return result_image
|
| 195 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
|
| 197 |
if USE_ZERO_GPU:
|
| 198 |
infer = spaces.GPU(infer)
|
|
|
|
| 199 |
|
| 200 |
if __name__ == "__main__":
|
| 201 |
-
demo =
|
| 202 |
demo.launch(server_name="0.0.0.0", ssr_mode=False)
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
import spaces
|
| 3 |
+
from ominicontrol import generate_image, generate_image_with_prompt
|
| 4 |
import os
|
| 5 |
|
| 6 |
from huggingface_hub import login
|
|
|
|
| 193 |
)
|
| 194 |
return result_image
|
| 195 |
|
| 196 |
+
def prompt_gradio_interface():
|
| 197 |
+
with gr.Blocks(css=css) as demo:
|
| 198 |
+
with gr.Row(equal_height=False):
|
| 199 |
+
with gr.Column(variant="panel", elem_classes="inputPanel"):
|
| 200 |
+
original_image = gr.Image(
|
| 201 |
+
type="pil",
|
| 202 |
+
label="Condition Image",
|
| 203 |
+
width=400,
|
| 204 |
+
height=400,
|
| 205 |
+
)
|
| 206 |
+
prompt = gr.Textbox(
|
| 207 |
+
label="Prompt",
|
| 208 |
+
)
|
| 209 |
+
# Advanced settings
|
| 210 |
+
with gr.Accordion(
|
| 211 |
+
"⚙️ Advanced Settings", open=False
|
| 212 |
+
) as advanced_settings:
|
| 213 |
+
inference_mode = gr.Radio(
|
| 214 |
+
["High Quality", "Fast"],
|
| 215 |
+
value="High Quality",
|
| 216 |
+
label="Generating Mode",
|
| 217 |
+
)
|
| 218 |
+
image_ratio = gr.Radio(
|
| 219 |
+
["Auto", "Square(1:1)", "Portrait(2:3)", "Landscape(3:2)"],
|
| 220 |
+
label="Image Ratio",
|
| 221 |
+
value="Auto",
|
| 222 |
+
)
|
| 223 |
+
use_random_seed = gr.Checkbox(label="Use Random Seed", value=True)
|
| 224 |
+
seed = gr.Number(
|
| 225 |
+
label="Seed",
|
| 226 |
+
value=42,
|
| 227 |
+
visible=(not use_random_seed.value),
|
| 228 |
+
)
|
| 229 |
+
use_random_seed.change(
|
| 230 |
+
lambda x: gr.update(visible=(not x)),
|
| 231 |
+
use_random_seed,
|
| 232 |
+
seed,
|
| 233 |
+
show_progress="hidden",
|
| 234 |
+
)
|
| 235 |
+
image_guidance = gr.Slider(
|
| 236 |
+
label="Image Guidance",
|
| 237 |
+
minimum=1.1,
|
| 238 |
+
maximum=5,
|
| 239 |
+
value=1.5,
|
| 240 |
+
step=0.1,
|
| 241 |
+
)
|
| 242 |
+
steps = gr.Slider(
|
| 243 |
+
label="Steps",
|
| 244 |
+
minimum=10,
|
| 245 |
+
maximum=50,
|
| 246 |
+
value=20,
|
| 247 |
+
step=1,
|
| 248 |
+
)
|
| 249 |
+
inference_mode.change(
|
| 250 |
+
lambda x: gr.update(interactive=(x == "High Quality")),
|
| 251 |
+
inference_mode,
|
| 252 |
+
image_guidance,
|
| 253 |
+
show_progress="hidden",
|
| 254 |
+
)
|
| 255 |
+
|
| 256 |
+
btn = gr.Button("Generate Image", variant="primary")
|
| 257 |
+
|
| 258 |
+
with gr.Column(elem_classes="outputPanel"):
|
| 259 |
+
output_image = gr.Image(
|
| 260 |
+
type="pil",
|
| 261 |
+
width=600,
|
| 262 |
+
height=600,
|
| 263 |
+
label="Output Image",
|
| 264 |
+
interactive=False,
|
| 265 |
+
sources=None,
|
| 266 |
+
)
|
| 267 |
+
|
| 268 |
+
# with gr.Row():
|
| 269 |
+
btn.click(
|
| 270 |
+
fn=prompt_infer,
|
| 271 |
+
inputs=[
|
| 272 |
+
original_image,
|
| 273 |
+
prompt,
|
| 274 |
+
inference_mode,
|
| 275 |
+
image_guidance,
|
| 276 |
+
image_ratio,
|
| 277 |
+
use_random_seed,
|
| 278 |
+
seed,
|
| 279 |
+
steps,
|
| 280 |
+
],
|
| 281 |
+
outputs=[
|
| 282 |
+
output_image,
|
| 283 |
+
],
|
| 284 |
+
)
|
| 285 |
+
|
| 286 |
+
return demo
|
| 287 |
+
|
| 288 |
+
def prompt_infer(
|
| 289 |
+
original_image,
|
| 290 |
+
prompt,
|
| 291 |
+
inference_mode,
|
| 292 |
+
image_guidance,
|
| 293 |
+
image_ratio,
|
| 294 |
+
use_random_seed,
|
| 295 |
+
seed,
|
| 296 |
+
steps,
|
| 297 |
+
):
|
| 298 |
+
result_image = generate_image_with_prompt(
|
| 299 |
+
image=original_image,
|
| 300 |
+
prompt=prompt,
|
| 301 |
+
inference_mode=inference_mode,
|
| 302 |
+
image_guidance=image_guidance,
|
| 303 |
+
image_ratio=image_ratio,
|
| 304 |
+
use_random_seed=use_random_seed,
|
| 305 |
+
seed=seed,
|
| 306 |
+
steps=steps,
|
| 307 |
+
)
|
| 308 |
+
return result_image
|
| 309 |
+
|
| 310 |
+
def multi_gradio_interface():
|
| 311 |
+
with gr.Blocks(css="style.css") as demo:
|
| 312 |
+
with gr.Tabs():
|
| 313 |
+
with gr.Tab(label="Style"):
|
| 314 |
+
gradio_interface()
|
| 315 |
+
with gr.Tab(label="Prompt"):
|
| 316 |
+
prompt_gradio_interface()
|
| 317 |
+
|
| 318 |
+
return demo
|
| 319 |
|
| 320 |
if USE_ZERO_GPU:
|
| 321 |
infer = spaces.GPU(infer)
|
| 322 |
+
prompt_infer = spaces.GPU(prompt_infer)
|
| 323 |
|
| 324 |
if __name__ == "__main__":
|
| 325 |
+
demo = multi_gradio_interface()
|
| 326 |
demo.launch(server_name="0.0.0.0", ssr_mode=False)
|
ominicontrol.py
CHANGED
|
@@ -12,6 +12,9 @@ pipe = FluxPipeline.from_pretrained(
|
|
| 12 |
)
|
| 13 |
pipe = pipe.to("cuda")
|
| 14 |
|
|
|
|
|
|
|
|
|
|
| 15 |
pipe.unload_lora_weights()
|
| 16 |
|
| 17 |
pipe.load_lora_weights(
|
|
@@ -34,19 +37,6 @@ pipe.load_lora_weights(
|
|
| 34 |
weight_name=f"v0/snoopy.safetensors",
|
| 35 |
adapter_name="snoopy",
|
| 36 |
)
|
| 37 |
-
# ref: https://civitai.com/models/715472/flux-hayao-miyazaki-ghibli
|
| 38 |
-
pipe.load_lora_weights(
|
| 39 |
-
"./lora",
|
| 40 |
-
weight_name="MaoMu_Ghibli.safetensors",
|
| 41 |
-
adapter_name="MaoMu_Ghibli",
|
| 42 |
-
)
|
| 43 |
-
# ref: https://civitai.com/models/824739/flux-3d-animation-style-lora
|
| 44 |
-
pipe.load_lora_weights(
|
| 45 |
-
"./lora",
|
| 46 |
-
weight_name="3d_animation.safetensors",
|
| 47 |
-
adapter_name="3d_animation",
|
| 48 |
-
)
|
| 49 |
-
|
| 50 |
|
| 51 |
def generate_image(
|
| 52 |
image,
|
|
@@ -72,8 +62,6 @@ def generate_image(
|
|
| 72 |
"Irasutoya Illustration": "irasutoya",
|
| 73 |
"The Simpsons": "simpsons",
|
| 74 |
"Snoopy": "snoopy",
|
| 75 |
-
"3D Animation": "3d_animation",
|
| 76 |
-
"MaoMu Ghibli": "MaoMu_Ghibli",
|
| 77 |
}[style]
|
| 78 |
pipe.set_adapters(activate_adapter_name)
|
| 79 |
|
|
@@ -145,3 +133,89 @@ def generate_image(
|
|
| 145 |
|
| 146 |
return result_img
|
| 147 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
)
|
| 13 |
pipe = pipe.to("cuda")
|
| 14 |
|
| 15 |
+
prompt_pipe = FluxPipeline.from_pipe(pipe)
|
| 16 |
+
prompt_pipe = prompt_pipe.to("cuda")
|
| 17 |
+
|
| 18 |
pipe.unload_lora_weights()
|
| 19 |
|
| 20 |
pipe.load_lora_weights(
|
|
|
|
| 37 |
weight_name=f"v0/snoopy.safetensors",
|
| 38 |
adapter_name="snoopy",
|
| 39 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 40 |
|
| 41 |
def generate_image(
|
| 42 |
image,
|
|
|
|
| 62 |
"Irasutoya Illustration": "irasutoya",
|
| 63 |
"The Simpsons": "simpsons",
|
| 64 |
"Snoopy": "snoopy",
|
|
|
|
|
|
|
| 65 |
}[style]
|
| 66 |
pipe.set_adapters(activate_adapter_name)
|
| 67 |
|
|
|
|
| 133 |
|
| 134 |
return result_img
|
| 135 |
|
| 136 |
+
|
| 137 |
+
def generate_image_with_prompt(
|
| 138 |
+
image,
|
| 139 |
+
prompt,
|
| 140 |
+
inference_mode,
|
| 141 |
+
image_guidance,
|
| 142 |
+
image_ratio,
|
| 143 |
+
steps,
|
| 144 |
+
use_random_seed,
|
| 145 |
+
seed,
|
| 146 |
+
):
|
| 147 |
+
# Prepare Condition
|
| 148 |
+
def resize(img, factor=16):
|
| 149 |
+
w, h = img.size
|
| 150 |
+
new_w, new_h = w // factor * factor, h // factor * factor
|
| 151 |
+
padding_w, padding_h = (w - new_w) // 2, (h - new_h) // 2
|
| 152 |
+
img = img.crop((padding_w, padding_h, new_w + padding_w, new_h + padding_h))
|
| 153 |
+
return img
|
| 154 |
+
|
| 155 |
+
original_width, original_height = image.size
|
| 156 |
+
|
| 157 |
+
factor = 512 / max(image.size)
|
| 158 |
+
image = resize(
|
| 159 |
+
image.resize(
|
| 160 |
+
(int(image.size[0] * factor), int(image.size[1] * factor)),
|
| 161 |
+
Image.LANCZOS,
|
| 162 |
+
)
|
| 163 |
+
)
|
| 164 |
+
|
| 165 |
+
delta = -image.size[0] // 16
|
| 166 |
+
condition = Condition(
|
| 167 |
+
"subject",
|
| 168 |
+
# activate_adapter_name,
|
| 169 |
+
image,
|
| 170 |
+
position_delta=(0, delta),
|
| 171 |
+
)
|
| 172 |
+
|
| 173 |
+
# Prepare seed
|
| 174 |
+
if use_random_seed:
|
| 175 |
+
seed = random.randint(0, 2**32 - 1)
|
| 176 |
+
seed_everything(seed)
|
| 177 |
+
|
| 178 |
+
# Image guidance scale
|
| 179 |
+
image_guidance = 1.0 if inference_mode == "Fast" else image_guidance
|
| 180 |
+
|
| 181 |
+
# Output size
|
| 182 |
+
if image_ratio == "Auto":
|
| 183 |
+
r = image.size[0] / image.size[1]
|
| 184 |
+
ratio = min([0.67, 1, 1.5], key=lambda x: abs(x - r))
|
| 185 |
+
else:
|
| 186 |
+
ratio = {
|
| 187 |
+
"Square(1:1)": 1,
|
| 188 |
+
"Portrait(2:3)": 0.67,
|
| 189 |
+
"Landscape(3:2)": 1.5,
|
| 190 |
+
}[image_ratio]
|
| 191 |
+
width, height = {
|
| 192 |
+
0.67: (640, 960),
|
| 193 |
+
1: (640, 640),
|
| 194 |
+
1.5: (960, 640),
|
| 195 |
+
}[ratio]
|
| 196 |
+
|
| 197 |
+
|
| 198 |
+
output_factor = max(width, height) / max(original_width, original_height)
|
| 199 |
+
width = int(original_width * output_factor)
|
| 200 |
+
height = int(original_height * output_factor)
|
| 201 |
+
|
| 202 |
+
print(
|
| 203 |
+
f"Image Ratio: {image_ratio}, Inference Mode: {inference_mode}, Image Guidance: {image_guidance}, Seed: {seed}, Steps: {steps}, Ratio: {ratio}, Size: {width}x{height}"
|
| 204 |
+
)
|
| 205 |
+
# Generate
|
| 206 |
+
result_img = generate(
|
| 207 |
+
prompt_pipe,
|
| 208 |
+
prompt=prompt,
|
| 209 |
+
conditions=[condition],
|
| 210 |
+
num_inference_steps=steps,
|
| 211 |
+
width=width,
|
| 212 |
+
height=height,
|
| 213 |
+
image_guidance_scale=image_guidance,
|
| 214 |
+
default_lora=True,
|
| 215 |
+
max_sequence_length=32,
|
| 216 |
+
).images[0]
|
| 217 |
+
# result_img = image
|
| 218 |
+
|
| 219 |
+
result_img = result_img.resize((width, height), Image.LANCZOS)
|
| 220 |
+
|
| 221 |
+
return result_img
|