Spaces:
Running
on
Zero
Running
on
Zero
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,317 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python
|
2 |
+
|
3 |
+
import os
|
4 |
+
import random
|
5 |
+
|
6 |
+
import gradio as gr
|
7 |
+
import numpy as np
|
8 |
+
import PIL.Image
|
9 |
+
import torch
|
10 |
+
import torchvision.transforms.functional as TF
|
11 |
+
|
12 |
+
from diffusers import ControlNetModel, StableDiffusionXLControlNetPipeline, AutoencoderKL
|
13 |
+
from diffusers import DDIMScheduler, EulerAncestralDiscreteScheduler
|
14 |
+
from controlnet_aux import PidiNetDetector, HEDdetector
|
15 |
+
from diffusers.utils import load_image
|
16 |
+
from huggingface_hub import HfApi
|
17 |
+
from pathlib import Path
|
18 |
+
from PIL import Image
|
19 |
+
import torch
|
20 |
+
import numpy as np
|
21 |
+
import cv2
|
22 |
+
import os
|
23 |
+
import random
|
24 |
+
|
25 |
+
|
26 |
+
def nms(x, t, s):
|
27 |
+
x = cv2.GaussianBlur(x.astype(np.float32), (0, 0), s)
|
28 |
+
|
29 |
+
f1 = np.array([[0, 0, 0], [1, 1, 1], [0, 0, 0]], dtype=np.uint8)
|
30 |
+
f2 = np.array([[0, 1, 0], [0, 1, 0], [0, 1, 0]], dtype=np.uint8)
|
31 |
+
f3 = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]], dtype=np.uint8)
|
32 |
+
f4 = np.array([[0, 0, 1], [0, 1, 0], [1, 0, 0]], dtype=np.uint8)
|
33 |
+
|
34 |
+
y = np.zeros_like(x)
|
35 |
+
|
36 |
+
for f in [f1, f2, f3, f4]:
|
37 |
+
np.putmask(y, cv2.dilate(x, kernel=f) == x, x)
|
38 |
+
|
39 |
+
z = np.zeros_like(y, dtype=np.uint8)
|
40 |
+
z[y > t] = 255
|
41 |
+
return z
|
42 |
+
|
43 |
+
|
44 |
+
DESCRIPTION = '''#
|
45 |
+
sketch to image with SDXL, using [@xinsir](https://huggingface.co/xinsir) [scribble sdxl controlnet](https://huggingface.co/xinsir/controlnet-scribble-sdxl-1.0)
|
46 |
+
'''
|
47 |
+
|
48 |
+
if not torch.cuda.is_available():
|
49 |
+
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
50 |
+
|
51 |
+
style_list = [
|
52 |
+
{
|
53 |
+
"name": "(No style)",
|
54 |
+
"prompt": "{prompt}",
|
55 |
+
"negative_prompt": "",
|
56 |
+
},
|
57 |
+
{
|
58 |
+
"name": "Cinematic",
|
59 |
+
"prompt": "cinematic still {prompt} . emotional, harmonious, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy",
|
60 |
+
"negative_prompt": "anime, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured",
|
61 |
+
},
|
62 |
+
{
|
63 |
+
"name": "3D Model",
|
64 |
+
"prompt": "professional 3d model {prompt} . octane render, highly detailed, volumetric, dramatic lighting",
|
65 |
+
"negative_prompt": "ugly, deformed, noisy, low poly, blurry, painting",
|
66 |
+
},
|
67 |
+
{
|
68 |
+
"name": "Anime",
|
69 |
+
"prompt": "anime artwork {prompt} . anime style, key visual, vibrant, studio anime, highly detailed",
|
70 |
+
"negative_prompt": "photo, deformed, black and white, realism, disfigured, low contrast",
|
71 |
+
},
|
72 |
+
{
|
73 |
+
"name": "Digital Art",
|
74 |
+
"prompt": "concept art {prompt} . digital artwork, illustrative, painterly, matte painting, highly detailed",
|
75 |
+
"negative_prompt": "photo, photorealistic, realism, ugly",
|
76 |
+
},
|
77 |
+
{
|
78 |
+
"name": "Photographic",
|
79 |
+
"prompt": "cinematic photo {prompt} . 35mm photograph, film, bokeh, professional, 4k, highly detailed",
|
80 |
+
"negative_prompt": "drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly",
|
81 |
+
},
|
82 |
+
{
|
83 |
+
"name": "Pixel art",
|
84 |
+
"prompt": "pixel-art {prompt} . low-res, blocky, pixel art style, 8-bit graphics",
|
85 |
+
"negative_prompt": "sloppy, messy, blurry, noisy, highly detailed, ultra textured, photo, realistic",
|
86 |
+
},
|
87 |
+
{
|
88 |
+
"name": "Fantasy art",
|
89 |
+
"prompt": "ethereal fantasy concept art of {prompt} . magnificent, celestial, ethereal, painterly, epic, majestic, magical, fantasy art, cover art, dreamy",
|
90 |
+
"negative_prompt": "photographic, realistic, realism, 35mm film, dslr, cropped, frame, text, deformed, glitch, noise, noisy, off-center, deformed, cross-eyed, closed eyes, bad anatomy, ugly, disfigured, sloppy, duplicate, mutated, black and white",
|
91 |
+
},
|
92 |
+
{
|
93 |
+
"name": "Neonpunk",
|
94 |
+
"prompt": "neonpunk style {prompt} . cyberpunk, vaporwave, neon, vibes, vibrant, stunningly beautiful, crisp, detailed, sleek, ultramodern, magenta highlights, dark purple shadows, high contrast, cinematic, ultra detailed, intricate, professional",
|
95 |
+
"negative_prompt": "painting, drawing, illustration, glitch, deformed, mutated, cross-eyed, ugly, disfigured",
|
96 |
+
},
|
97 |
+
{
|
98 |
+
"name": "Manga",
|
99 |
+
"prompt": "manga style {prompt} . vibrant, high-energy, detailed, iconic, Japanese comic style",
|
100 |
+
"negative_prompt": "ugly, deformed, noisy, blurry, low contrast, realism, photorealistic, Western comic style",
|
101 |
+
},
|
102 |
+
]
|
103 |
+
|
104 |
+
styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list}
|
105 |
+
STYLE_NAMES = list(styles.keys())
|
106 |
+
DEFAULT_STYLE_NAME = "(No style)"
|
107 |
+
|
108 |
+
|
109 |
+
def apply_style(style_name: str, positive: str, negative: str = "") -> tuple[str, str]:
|
110 |
+
p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
|
111 |
+
return p.replace("{prompt}", positive), n + negative
|
112 |
+
|
113 |
+
|
114 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
115 |
+
|
116 |
+
eulera_scheduler = EulerAncestralDiscreteScheduler.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", subfolder="scheduler")
|
117 |
+
|
118 |
+
|
119 |
+
controlnet = ControlNetModel.from_pretrained(
|
120 |
+
"xinsir/controlnet-scribble-sdxl-1.0",
|
121 |
+
torch_dtype=torch.float16
|
122 |
+
)
|
123 |
+
|
124 |
+
# when test with other base model, you need to change the vae also.
|
125 |
+
vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
|
126 |
+
|
127 |
+
pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
|
128 |
+
"stabilityai/stable-diffusion-xl-base-1.0",
|
129 |
+
controlnet=controlnet,
|
130 |
+
vae=vae,
|
131 |
+
torch_dtype=torch.float16,
|
132 |
+
scheduler=eulera_scheduler,
|
133 |
+
)
|
134 |
+
pipe.to(device)
|
135 |
+
# Load model.
|
136 |
+
|
137 |
+
MAX_SEED = np.iinfo(np.int32).max
|
138 |
+
processor = HEDdetector.from_pretrained('lllyasviel/Annotators')
|
139 |
+
def nms(x, t, s):
|
140 |
+
x = cv2.GaussianBlur(x.astype(np.float32), (0, 0), s)
|
141 |
+
|
142 |
+
f1 = np.array([[0, 0, 0], [1, 1, 1], [0, 0, 0]], dtype=np.uint8)
|
143 |
+
f2 = np.array([[0, 1, 0], [0, 1, 0], [0, 1, 0]], dtype=np.uint8)
|
144 |
+
f3 = np.array([[1, 0, 0], [0, 1, 0], [0, 0, 1]], dtype=np.uint8)
|
145 |
+
f4 = np.array([[0, 0, 1], [0, 1, 0], [1, 0, 0]], dtype=np.uint8)
|
146 |
+
|
147 |
+
y = np.zeros_like(x)
|
148 |
+
|
149 |
+
for f in [f1, f2, f3, f4]:
|
150 |
+
np.putmask(y, cv2.dilate(x, kernel=f) == x, x)
|
151 |
+
|
152 |
+
z = np.zeros_like(y, dtype=np.uint8)
|
153 |
+
z[y > t] = 255
|
154 |
+
return z
|
155 |
+
|
156 |
+
def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
|
157 |
+
if randomize_seed:
|
158 |
+
seed = random.randint(0, MAX_SEED)
|
159 |
+
return seed
|
160 |
+
|
161 |
+
|
162 |
+
def run(
|
163 |
+
image: PIL.Image.Image,
|
164 |
+
prompt: str,
|
165 |
+
negative_prompt: str,
|
166 |
+
style_name: str = DEFAULT_STYLE_NAME,
|
167 |
+
num_steps: int = 25,
|
168 |
+
guidance_scale: float = 5,
|
169 |
+
controlnet_conditioning_scale: float = 1.0,
|
170 |
+
seed: int = 0,
|
171 |
+
use_hed: bool = False,
|
172 |
+
progress=gr.Progress(track_tqdm=True),
|
173 |
+
) -> PIL.Image.Image:
|
174 |
+
# image = image.convert("RGB")
|
175 |
+
# image = TF.to_tensor(image) > 0.5
|
176 |
+
# image = TF.to_pil_image(image.to(torch.float32))
|
177 |
+
width, height = image['composite'].size
|
178 |
+
ratio = np.sqrt(1024. * 1024. / (width * height))
|
179 |
+
new_width, new_height = int(width * ratio), int(height * ratio)
|
180 |
+
image = image['composite'].resize((new_width, new_height))
|
181 |
+
|
182 |
+
if use_hed:
|
183 |
+
controlnet_img = processor(image, scribble=False)
|
184 |
+
# following is some processing to simulate human sketch draw, different threshold can generate different width of lines
|
185 |
+
controlnet_img = np.array(controlnet_img)
|
186 |
+
controlnet_img = nms(controlnet_img, 127, 3)
|
187 |
+
controlnet_img = cv2.GaussianBlur(controlnet_img, (0, 0), 3)
|
188 |
+
|
189 |
+
# higher threshold, thiner line
|
190 |
+
random_val = int(round(random.uniform(0.01, 0.10), 2) * 255)
|
191 |
+
controlnet_img[controlnet_img > random_val] = 255
|
192 |
+
controlnet_img[controlnet_img < 255] = 0
|
193 |
+
image = Image.fromarray(controlnet_img)
|
194 |
+
|
195 |
+
|
196 |
+
|
197 |
+
prompt, negative_prompt = apply_style(style_name, prompt, negative_prompt)
|
198 |
+
|
199 |
+
generator = torch.Generator(device=device).manual_seed(seed)
|
200 |
+
out = pipe(
|
201 |
+
prompt=prompt,
|
202 |
+
negative_prompt=negative_prompt,
|
203 |
+
image=image,
|
204 |
+
num_inference_steps=num_steps,
|
205 |
+
generator=generator,
|
206 |
+
controlnet_conditioning_scale=controlnet_conditioning_scale,
|
207 |
+
guidance_scale=guidance_scale,
|
208 |
+
width=new_width,
|
209 |
+
height=new_height,
|
210 |
+
).images[0]
|
211 |
+
|
212 |
+
return out
|
213 |
+
|
214 |
+
|
215 |
+
with gr.Blocks(css="style.css") as demo:
|
216 |
+
gr.Markdown(DESCRIPTION, elem_id="description")
|
217 |
+
gr.DuplicateButton(
|
218 |
+
value="Duplicate Space for private use",
|
219 |
+
elem_id="duplicate-button",
|
220 |
+
visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
|
221 |
+
)
|
222 |
+
|
223 |
+
with gr.Row():
|
224 |
+
with gr.Column():
|
225 |
+
with gr.Group():
|
226 |
+
image = gr.ImageEditor(type="pil", image_mode="L", crop_size=(512, 512),brush=gr.Brush(color_mode="fixed", colors=["#00000"]))
|
227 |
+
prompt = gr.Textbox(label="Prompt")
|
228 |
+
style = gr.Dropdown(label="Style", choices=STYLE_NAMES, value=DEFAULT_STYLE_NAME)
|
229 |
+
use_hed = gr.Checkbox(label="use HED detector", value=False)
|
230 |
+
run_button = gr.Button("Run")
|
231 |
+
with gr.Accordion("Advanced options", open=False):
|
232 |
+
negative_prompt = gr.Textbox(
|
233 |
+
label="Negative prompt",
|
234 |
+
value=" extra digit, fewer digits, cropped, worst quality, low quality, glitch, deformed, mutated, ugly, disfigured",
|
235 |
+
)
|
236 |
+
num_steps = gr.Slider(
|
237 |
+
label="Number of steps",
|
238 |
+
minimum=1,
|
239 |
+
maximum=50,
|
240 |
+
step=1,
|
241 |
+
value=25,
|
242 |
+
)
|
243 |
+
guidance_scale = gr.Slider(
|
244 |
+
label="Guidance scale",
|
245 |
+
minimum=0.1,
|
246 |
+
maximum=10.0,
|
247 |
+
step=0.1,
|
248 |
+
value=5,
|
249 |
+
)
|
250 |
+
controlnet_conditioning_scale = gr.Slider(
|
251 |
+
label="controlnet conditioning scale",
|
252 |
+
minimum=0.5,
|
253 |
+
maximum=5.0,
|
254 |
+
step=0.1,
|
255 |
+
value=0.9,
|
256 |
+
)
|
257 |
+
seed = gr.Slider(
|
258 |
+
label="Seed",
|
259 |
+
minimum=0,
|
260 |
+
maximum=MAX_SEED,
|
261 |
+
step=1,
|
262 |
+
value=0,
|
263 |
+
)
|
264 |
+
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
265 |
+
|
266 |
+
with gr.Column():
|
267 |
+
result = gr.Image(label="Result", height=400)
|
268 |
+
|
269 |
+
inputs = [
|
270 |
+
image,
|
271 |
+
prompt,
|
272 |
+
negative_prompt,
|
273 |
+
style,
|
274 |
+
num_steps,
|
275 |
+
guidance_scale,
|
276 |
+
controlnet_conditioning_scale,
|
277 |
+
seed,
|
278 |
+
use_hed,
|
279 |
+
]
|
280 |
+
prompt.submit(
|
281 |
+
fn=randomize_seed_fn,
|
282 |
+
inputs=[seed, randomize_seed],
|
283 |
+
outputs=seed,
|
284 |
+
queue=False,
|
285 |
+
api_name=False,
|
286 |
+
).then(
|
287 |
+
fn=run,
|
288 |
+
inputs=inputs,
|
289 |
+
outputs=result,
|
290 |
+
api_name=False,
|
291 |
+
)
|
292 |
+
negative_prompt.submit(
|
293 |
+
fn=randomize_seed_fn,
|
294 |
+
inputs=[seed, randomize_seed],
|
295 |
+
outputs=seed,
|
296 |
+
queue=False,
|
297 |
+
api_name=False,
|
298 |
+
).then(
|
299 |
+
fn=run,
|
300 |
+
inputs=inputs,
|
301 |
+
outputs=result,
|
302 |
+
api_name=False,
|
303 |
+
)
|
304 |
+
run_button.click(
|
305 |
+
fn=randomize_seed_fn,
|
306 |
+
inputs=[seed, randomize_seed],
|
307 |
+
outputs=seed,
|
308 |
+
queue=False,
|
309 |
+
api_name=False,
|
310 |
+
).then(
|
311 |
+
fn=run,
|
312 |
+
inputs=inputs,
|
313 |
+
outputs=result,
|
314 |
+
api_name=False,
|
315 |
+
)
|
316 |
+
|
317 |
+
demo.queue().launch()
|