Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -34,7 +34,6 @@ function refresh() {
|
|
| 34 |
}
|
| 35 |
}
|
| 36 |
"""
|
| 37 |
-
|
| 38 |
def nms(x, t, s):
|
| 39 |
x = cv2.GaussianBlur(x.astype(np.float32), (0, 0), s)
|
| 40 |
|
|
@@ -134,14 +133,17 @@ styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list}
|
|
| 134 |
STYLE_NAMES = list(styles.keys())
|
| 135 |
DEFAULT_STYLE_NAME = "(No style)"
|
| 136 |
|
|
|
|
| 137 |
def apply_style(style_name: str, positive: str, negative: str = "") -> tuple[str, str]:
|
| 138 |
p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
|
| 139 |
return p.replace("{prompt}", positive), n + negative
|
| 140 |
|
|
|
|
| 141 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 142 |
|
| 143 |
eulera_scheduler = EulerAncestralDiscreteScheduler.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", subfolder="scheduler")
|
| 144 |
|
|
|
|
| 145 |
controlnet = ControlNetModel.from_pretrained(
|
| 146 |
"xinsir/controlnet-scribble-sdxl-1.0",
|
| 147 |
torch_dtype=torch.float16
|
|
@@ -173,98 +175,181 @@ pipe_canny = StableDiffusionXLControlNetPipeline.from_pretrained(
|
|
| 173 |
|
| 174 |
pipe_canny.to(device)
|
| 175 |
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
def
|
| 179 |
-
|
| 180 |
-
width, height = image.size
|
| 181 |
-
if width * height > max_pixels:
|
| 182 |
-
scale_factor = (max_pixels / (width * height)) ** 0.5
|
| 183 |
-
new_size = (int(width * scale_factor), int(height * scale_factor))
|
| 184 |
-
return image.resize(new_size, Image.ANTIALIAS)
|
| 185 |
-
return image
|
| 186 |
-
|
| 187 |
-
def process(image, prompt, style, detector_name):
|
| 188 |
-
# Convert image to RGB mode if it's not already
|
| 189 |
-
if image.mode != 'RGB':
|
| 190 |
-
image = image.convert('RGB')
|
| 191 |
-
image = resize_image(image)
|
| 192 |
-
|
| 193 |
-
width, height = image.size
|
| 194 |
-
|
| 195 |
-
prompt, negative_prompt = apply_style(style, prompt)
|
| 196 |
-
|
| 197 |
-
if detector_name == "hed":
|
| 198 |
-
image = HWC3(np.array(image, dtype=np.uint8))
|
| 199 |
-
with torch.no_grad():
|
| 200 |
-
detected_map = hed(image, scribble=True)
|
| 201 |
-
detected_map = HWC3(detected_map)
|
| 202 |
-
image = Image.fromarray(detected_map)
|
| 203 |
-
images = pipe(prompt, negative_prompt=negative_prompt, image=image, height=height, width=width).images
|
| 204 |
-
return images[0]
|
| 205 |
-
elif detector_name == "scribble":
|
| 206 |
-
image = HWC3(np.array(image, dtype=np.uint8))
|
| 207 |
-
with torch.no_grad():
|
| 208 |
-
detected_map = nms(image, 127, 3.0)
|
| 209 |
-
detected_map = HWC3(detected_map)
|
| 210 |
-
image = Image.fromarray(detected_map)
|
| 211 |
-
images = pipe(prompt, negative_prompt=negative_prompt, image=image, height=height, width=width).images
|
| 212 |
-
return images[0]
|
| 213 |
-
elif detector_name == "canny":
|
| 214 |
-
image = np.array(image, dtype=np.uint8)
|
| 215 |
-
image = cv2.Canny(image, 100, 200)
|
| 216 |
-
image = image[:, :, None]
|
| 217 |
-
image = np.concatenate([image, image, image], axis=2)
|
| 218 |
-
detected_map = image
|
| 219 |
-
image = Image.fromarray(detected_map)
|
| 220 |
-
images = pipe_canny(prompt, negative_prompt=negative_prompt, image=image, height=height, width=width).images
|
| 221 |
-
return images[0]
|
| 222 |
-
|
| 223 |
-
block_css = (
|
| 224 |
-
code := """
|
| 225 |
-
#image_upload {
|
| 226 |
-
height: 100% !important;
|
| 227 |
-
}
|
| 228 |
-
#prompt_input {
|
| 229 |
-
height: 100% !important;
|
| 230 |
-
}
|
| 231 |
-
#select_style {
|
| 232 |
-
height: 100% !important;
|
| 233 |
-
}
|
| 234 |
-
#detect_method {
|
| 235 |
-
height: 100% !important;
|
| 236 |
-
}
|
| 237 |
-
#submit_button {
|
| 238 |
-
height: 100% !important;
|
| 239 |
-
}
|
| 240 |
-
"""
|
| 241 |
-
)
|
| 242 |
|
| 243 |
-
|
| 244 |
-
|
| 245 |
-
|
| 246 |
-
|
| 247 |
-
|
| 248 |
-
|
| 249 |
-
|
| 250 |
-
|
| 251 |
-
|
| 252 |
-
|
| 253 |
-
|
| 254 |
-
|
| 255 |
-
|
| 256 |
-
gallery = gr.Gallery(label="Generated images", show_label=False, elem_id="gallery").style(grid=2, height="auto")
|
| 257 |
-
|
| 258 |
-
submit_btn.click(process, inputs=[input_image, prompt, style, detect_method], outputs=[gallery])
|
| 259 |
-
|
| 260 |
-
# Refresh button to apply the dark theme
|
| 261 |
-
refresh_btn = gr.Button("Refresh for Dark Theme")
|
| 262 |
-
refresh_btn.click(None, None, None, _js=js_func)
|
| 263 |
-
|
| 264 |
-
return demo
|
| 265 |
|
| 266 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 267 |
|
| 268 |
-
|
| 269 |
-
demo = create_demo()
|
| 270 |
-
demo.launch(debug=True)
|
|
|
|
| 34 |
}
|
| 35 |
}
|
| 36 |
"""
|
|
|
|
| 37 |
def nms(x, t, s):
|
| 38 |
x = cv2.GaussianBlur(x.astype(np.float32), (0, 0), s)
|
| 39 |
|
|
|
|
| 133 |
STYLE_NAMES = list(styles.keys())
|
| 134 |
DEFAULT_STYLE_NAME = "(No style)"
|
| 135 |
|
| 136 |
+
|
| 137 |
def apply_style(style_name: str, positive: str, negative: str = "") -> tuple[str, str]:
|
| 138 |
p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
|
| 139 |
return p.replace("{prompt}", positive), n + negative
|
| 140 |
|
| 141 |
+
|
| 142 |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
| 143 |
|
| 144 |
eulera_scheduler = EulerAncestralDiscreteScheduler.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", subfolder="scheduler")
|
| 145 |
|
| 146 |
+
|
| 147 |
controlnet = ControlNetModel.from_pretrained(
|
| 148 |
"xinsir/controlnet-scribble-sdxl-1.0",
|
| 149 |
torch_dtype=torch.float16
|
|
|
|
| 175 |
|
| 176 |
pipe_canny.to(device)
|
| 177 |
|
| 178 |
+
MAX_SEED = np.iinfo(np.int32).max
|
| 179 |
+
processor = HEDdetector.from_pretrained('lllyasviel/Annotators')
|
| 180 |
+
def nms(x, t, s):
|
| 181 |
+
x = cv2.GaussianBlur(x.astype(np.float32), (0, 0), s)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 182 |
|
| 183 |
+
f1 = np.array([[0, 0, 0], [1, 1, 1], [0, 0, 0]], dtype=np.uint8)
|
| 184 |
+
f2 = np.array([[0, 1, 0], [0, 1, 0], [0, 1, 0]], dtype=np.uint8)
|
| 185 |
+
f3 = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]], dtype=np.uint8)
|
| 186 |
+
f4 = np.array([[0, 0, 1], [0, 1, 0], [1, 0, 0]], dtype=np.uint8)
|
| 187 |
+
|
| 188 |
+
y = np.zeros_like(x)
|
| 189 |
+
|
| 190 |
+
for f in [f1, f2, f3, f4]:
|
| 191 |
+
np.putmask(y, cv2.dilate(x, kernel=f) == x, x)
|
| 192 |
+
|
| 193 |
+
z = np.zeros_like(y, dtype=np.uint8)
|
| 194 |
+
z[y > t] = 255
|
| 195 |
+
return z
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
|
| 197 |
+
def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
|
| 198 |
+
if randomize_seed:
|
| 199 |
+
seed = random.randint(0, MAX_SEED)
|
| 200 |
+
return seed
|
| 201 |
+
|
| 202 |
+
@spaces.GPU
|
| 203 |
+
def run(
|
| 204 |
+
image: PIL.Image.Image,
|
| 205 |
+
prompt: str,
|
| 206 |
+
negative_prompt: str,
|
| 207 |
+
style_name: str = DEFAULT_STYLE_NAME,
|
| 208 |
+
num_steps: int = 25,
|
| 209 |
+
guidance_scale: float = 5,
|
| 210 |
+
controlnet_conditioning_scale: float = 1.0,
|
| 211 |
+
seed: int = 0,
|
| 212 |
+
use_hed: bool = False,
|
| 213 |
+
use_canny: bool = False,
|
| 214 |
+
progress=gr.Progress(track_tqdm=True),
|
| 215 |
+
) -> PIL.Image.Image:
|
| 216 |
+
width, height = image['composite'].size
|
| 217 |
+
ratio = np.sqrt(1024. * 1024. / (width * height))
|
| 218 |
+
new_width, new_height = int(width * ratio), int(height * ratio)
|
| 219 |
+
image = image['composite'].resize((new_width, new_height))
|
| 220 |
+
|
| 221 |
+
if use_canny:
|
| 222 |
+
controlnet_img = np.array(image)
|
| 223 |
+
controlnet_img = cv2.Canny(controlnet_img, 100, 200)
|
| 224 |
+
controlnet_img = HWC3(controlnet_img)
|
| 225 |
+
image = Image.fromarray(controlnet_img)
|
| 226 |
+
|
| 227 |
+
elif not use_hed:
|
| 228 |
+
controlnet_img = image
|
| 229 |
+
else:
|
| 230 |
+
controlnet_img = processor(image, scribble=False)
|
| 231 |
+
# following is some processing to simulate human sketch draw, different threshold can generate different width of lines
|
| 232 |
+
controlnet_img = np.array(controlnet_img)
|
| 233 |
+
controlnet_img = nms(controlnet_img, 127, 3)
|
| 234 |
+
controlnet_img = cv2.GaussianBlur(controlnet_img, (0, 0), 3)
|
| 235 |
+
|
| 236 |
+
# higher threshold, thiner line
|
| 237 |
+
random_val = int(round(random.uniform(0.01, 0.10), 2) * 255)
|
| 238 |
+
controlnet_img[controlnet_img > random_val] = 255
|
| 239 |
+
controlnet_img[controlnet_img < 255] = 0
|
| 240 |
+
image = Image.fromarray(controlnet_img)
|
| 241 |
+
|
| 242 |
+
|
| 243 |
+
prompt, negative_prompt = apply_style(style_name, prompt, negative_prompt)
|
| 244 |
+
|
| 245 |
+
generator = torch.Generator(device=device).manual_seed(seed)
|
| 246 |
+
if use_canny:
|
| 247 |
+
out = pipe_canny(
|
| 248 |
+
prompt=prompt,
|
| 249 |
+
negative_prompt=negative_prompt,
|
| 250 |
+
image=image,
|
| 251 |
+
num_inference_steps=num_steps,
|
| 252 |
+
generator=generator,
|
| 253 |
+
controlnet_conditioning_scale=controlnet_conditioning_scale,
|
| 254 |
+
guidance_scale=guidance_scale,
|
| 255 |
+
width=new_width,
|
| 256 |
+
height=new_height,
|
| 257 |
+
).images[0]
|
| 258 |
+
else:
|
| 259 |
+
out = pipe(
|
| 260 |
+
prompt=prompt,
|
| 261 |
+
negative_prompt=negative_prompt,
|
| 262 |
+
image=image,
|
| 263 |
+
num_inference_steps=num_steps,
|
| 264 |
+
generator=generator,
|
| 265 |
+
controlnet_conditioning_scale=controlnet_conditioning_scale,
|
| 266 |
+
guidance_scale=guidance_scale,
|
| 267 |
+
width=new_width,
|
| 268 |
+
height=new_height,).images[0]
|
| 269 |
+
|
| 270 |
+
return (controlnet_img, out)
|
| 271 |
+
|
| 272 |
+
|
| 273 |
+
with gr.Blocks(css="style.css", js=js_func) as demo:
|
| 274 |
+
gr.Markdown(DESCRIPTION, elem_id="description")
|
| 275 |
+
gr.DuplicateButton(
|
| 276 |
+
value="Duplicate Space for private use",
|
| 277 |
+
elem_id="duplicate-button",
|
| 278 |
+
visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
|
| 279 |
+
)
|
| 280 |
+
|
| 281 |
+
with gr.Row():
|
| 282 |
+
with gr.Column():
|
| 283 |
+
with gr.Group():
|
| 284 |
+
image = gr.ImageEditor(type="pil", image_mode="L", crop_size=(512, 512))
|
| 285 |
+
prompt = gr.Textbox(label="Prompt")
|
| 286 |
+
style = gr.Dropdown(label="Style", choices=STYLE_NAMES, value=DEFAULT_STYLE_NAME)
|
| 287 |
+
use_hed = gr.Checkbox(label="use HED detector", value=False, info="check this box if you upload an image and want to turn it to a sketch")
|
| 288 |
+
use_canny = gr.Checkbox(label="use Canny", value=False, info="check this to use ControlNet canny instead of scribble")
|
| 289 |
+
run_button = gr.Button("Run")
|
| 290 |
+
with gr.Accordion("Advanced options", open=False):
|
| 291 |
+
negative_prompt = gr.Textbox(
|
| 292 |
+
label="Negative prompt",
|
| 293 |
+
value="longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality",
|
| 294 |
+
)
|
| 295 |
+
num_steps = gr.Slider(
|
| 296 |
+
label="Number of steps",
|
| 297 |
+
minimum=1,
|
| 298 |
+
maximum=50,
|
| 299 |
+
step=1,
|
| 300 |
+
value=25,
|
| 301 |
+
)
|
| 302 |
+
guidance_scale = gr.Slider(
|
| 303 |
+
label="Guidance scale",
|
| 304 |
+
minimum=0.1,
|
| 305 |
+
maximum=10.0,
|
| 306 |
+
step=0.1,
|
| 307 |
+
value=5,
|
| 308 |
+
)
|
| 309 |
+
controlnet_conditioning_scale = gr.Slider(
|
| 310 |
+
label="controlnet conditioning scale",
|
| 311 |
+
minimum=0.5,
|
| 312 |
+
maximum=5.0,
|
| 313 |
+
step=0.1,
|
| 314 |
+
value=0.9,
|
| 315 |
+
)
|
| 316 |
+
seed = gr.Slider(
|
| 317 |
+
label="Seed",
|
| 318 |
+
minimum=0,
|
| 319 |
+
maximum=MAX_SEED,
|
| 320 |
+
step=1,
|
| 321 |
+
value=0,
|
| 322 |
+
)
|
| 323 |
+
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
| 324 |
+
|
| 325 |
+
with gr.Column():
|
| 326 |
+
with gr.Group():
|
| 327 |
+
image_slider = ImageSlider(position=0.5)
|
| 328 |
+
|
| 329 |
+
|
| 330 |
+
inputs = [
|
| 331 |
+
image,
|
| 332 |
+
prompt,
|
| 333 |
+
negative_prompt,
|
| 334 |
+
style,
|
| 335 |
+
num_steps,
|
| 336 |
+
guidance_scale,
|
| 337 |
+
controlnet_conditioning_scale,
|
| 338 |
+
seed,
|
| 339 |
+
use_hed,
|
| 340 |
+
use_canny
|
| 341 |
+
]
|
| 342 |
+
outputs = [image_slider]
|
| 343 |
+
run_button.click(
|
| 344 |
+
fn=randomize_seed_fn,
|
| 345 |
+
inputs=[seed, randomize_seed],
|
| 346 |
+
outputs=seed,
|
| 347 |
+
queue=False,
|
| 348 |
+
api_name=False,
|
| 349 |
+
).then(lambda x: None, inputs=None, outputs=image_slider).then(
|
| 350 |
+
fn=run, inputs=inputs, outputs=outputs
|
| 351 |
+
)
|
| 352 |
+
|
| 353 |
+
|
| 354 |
|
| 355 |
+
demo.queue().launch()
|
|
|
|
|
|