Spaces:
Runtime error
Runtime error
File size: 14,873 Bytes
96d7ad8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 |
import os
import time
import pdb
import cuid
import gradio as gr
import spaces
import numpy as np
import sys
from huggingface_hub import snapshot_download
import subprocess
ProjectDir = os.path.abspath(os.path.dirname(__file__))
CheckpointsDir = os.path.join(ProjectDir, "checkpoints")
sys.path.insert(0, ProjectDir)
sys.path.insert(0, f"{ProjectDir}/MMCM")
sys.path.insert(0, f"{ProjectDir}/diffusers/src")
sys.path.insert(0, f"{ProjectDir}/controlnet_aux/src")
sys.path.insert(0, f"{ProjectDir}/scripts/gradio")
result = subprocess.run(
["pip", "install", "--no-cache-dir", "-U", "openmim"],
capture_output=True,
text=True,
)
print(result)
result = subprocess.run(["mim", "install", "mmengine"], capture_output=True, text=True)
print(result)
result = subprocess.run(
["mim", "install", "mmcv>=2.0.1"], capture_output=True, text=True
)
print(result)
result = subprocess.run(
["mim", "install", "mmdet>=3.1.0"], capture_output=True, text=True
)
print(result)
result = subprocess.run(
["mim", "install", "mmpose>=1.1.0"], capture_output=True, text=True
)
print(result)
ignore_video2video = True
max_image_edge = 960
def download_model():
if not os.path.exists(CheckpointsDir):
print("Checkpoint Not Downloaded, start downloading...")
tic = time.time()
snapshot_download(
repo_id="TMElyralab/MuseV",
local_dir=CheckpointsDir,
max_workers=8,
local_dir_use_symlinks=True,
)
toc = time.time()
print(f"download cost {toc-tic} seconds")
else:
print("Already download the model.")
download_model() # for huggingface deployment.
if not ignore_video2video:
from gradio_video2video import online_v2v_inference
from gradio_text2video import online_t2v_inference
@spaces.GPU(duration=180)
def hf_online_t2v_inference(
prompt,
image_np,
seed,
fps,
w,
h,
video_len,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_t2v_inference(
prompt, image_np, seed, fps, w, h, video_len, img_edge_ratio
)
@spaces.GPU(duration=180)
def hg_online_v2v_inference(
prompt,
image_np,
video,
processor,
seed,
fps,
w,
h,
video_length,
img_edge_ratio,
):
img_edge_ratio, _, _ = limit_shape(
image_np, w, h, img_edge_ratio, max_image_edge=max_image_edge
)
if not isinstance(image_np, np.ndarray): # None
raise gr.Error("Need input reference image")
return online_v2v_inference(
prompt,
image_np,
video,
processor,
seed,
fps,
w,
h,
video_length,
img_edge_ratio,
)
def limit_shape(image, input_w, input_h, img_edge_ratio, max_image_edge=max_image_edge):
"""limite generation video shape to avoid gpu memory overflow"""
if input_h == -1 and input_w == -1:
if isinstance(image, np.ndarray):
input_h, input_w, _ = image.shape
elif isinstance(image, PIL.Image.Image):
input_w, input_h = image.size
else:
raise ValueError(
f"image should be in [image, ndarray], but given {type(image)}"
)
if img_edge_ratio == 0:
img_edge_ratio = 1
img_edge_ratio_infact = min(max_image_edge / max(input_h, input_w), img_edge_ratio)
# print(
# image.shape,
# input_w,
# input_h,
# img_edge_ratio,
# max_image_edge,
# img_edge_ratio_infact,
# )
if img_edge_ratio != 1:
return (
img_edge_ratio_infact,
input_w * img_edge_ratio_infact,
input_h * img_edge_ratio_infact,
)
else:
return img_edge_ratio_infact, -1, -1
def limit_length(length):
"""limite generation video frames numer to avoid gpu memory overflow"""
if length > 24 * 6:
gr.Warning("Length need to smaller than 144, dute to gpu memory limit")
length = 24 * 6
return length
class ConcatenateBlock(gr.blocks.Block):
def __init__(self, options):
self.options = options
self.current_string = ""
def update_string(self, new_choice):
if new_choice and new_choice not in self.current_string.split(", "):
if self.current_string == "":
self.current_string = new_choice
else:
self.current_string += ", " + new_choice
return self.current_string
def process_input(new_choice):
return concatenate_block.update_string(new_choice), ""
control_options = [
"pose",
"pose_body",
"pose_hand",
"pose_face",
"pose_hand_body",
"pose_hand_face",
"dwpose",
"dwpose_face",
"dwpose_hand",
"dwpose_body",
"dwpose_body_hand",
"canny",
"tile",
"hed",
"hed_scribble",
"depth",
"pidi",
"normal_bae",
"lineart",
"lineart_anime",
"zoe",
"sam",
"mobile_sam",
"leres",
"content",
"face_detector",
]
concatenate_block = ConcatenateBlock(control_options)
css = """#input_img {max-width: 1024px !important} #output_vid {max-width: 1024px; max-height: 576px}"""
with gr.Blocks(css=css) as demo:
gr.Markdown(
"<div align='center'> <h1> MuseV: Infinite-length and High Fidelity Virtual Human Video Generation with Visual Conditioned Parallel Denoising</span> </h1> \
<h2 style='font-weight: 450; font-size: 1rem; margin: 0rem'>\
</br>\
Zhiqiang Xia <sup>*</sup>,\
Zhaokang Chen<sup>*</sup>,\
Bin Wu<sup>†</sup>,\
Chao Li,\
Kwok-Wai Hung,\
Chao Zhan,\
Yingjie He,\
Wenjiang Zhou\
(<sup>*</sup>Equal Contribution, <sup>†</sup>Corresponding Author, [email protected])\
</br>\
Lyra Lab, Tencent Music Entertainment\
</h2> \
<a style='font-size:18px;color: #000000' href='https://github.com/TMElyralab/MuseV'>[Github Repo]</a>\
<a style='font-size:18px;color: #000000'>, which is important to Open-Source projects. Thanks!</a>\
<a style='font-size:18px;color: #000000' href=''> [ArXiv(Coming Soon)] </a>\
<a style='font-size:18px;color: #000000' href=''> [Project Page(Coming Soon)] </a> \
<a style='font-size:18px;color: #000000'>If MuseV is useful, please help star the repo~ </a> </div>"
)
with gr.Tab("Text to Video"):
with gr.Row():
with gr.Column():
prompt = gr.Textbox(label="Prompt")
image = gr.Image(label="VisionCondImage")
seed = gr.Number(
label="Seed (seed=-1 means that the seeds run each time are different)",
value=-1,
)
video_length = gr.Number(
label="Video Length(need smaller than 144,If you want to be able to generate longer videos, run it locally )",
value=12,
)
fps = gr.Number(label="Generate Video FPS", value=6)
gr.Markdown(
(
"If W&H is -1, then use the Reference Image's Size. Size of target video is $(W, H)*img\_edge\_ratio$. \n"
"The shorter the image size, the larger the motion amplitude, and the lower video quality.\n"
"The longer the W&H, the smaller the motion amplitude, and the higher video quality.\n"
"Due to the GPU VRAM limits, the W&H need smaller than 960px"
)
)
with gr.Row():
w = gr.Number(label="Width", value=-1)
h = gr.Number(label="Height", value=-1)
img_edge_ratio = gr.Number(label="img_edge_ratio", value=1.0)
with gr.Row():
out_w = gr.Number(label="Output Width", value=0, interactive=False)
out_h = gr.Number(label="Output Height", value=0, interactive=False)
img_edge_ratio_infact = gr.Number(
label="img_edge_ratio in fact",
value=1.0,
interactive=False,
)
btn1 = gr.Button("Generate")
out = gr.Video()
# pdb.set_trace()
i2v_examples_256 = [
[
"(masterpiece, best quality, highres:1),(1boy, solo:1),(eye blinks:1.8),(head wave:1.3)",
"../../data/images/yongen.jpeg",
],
[
"(masterpiece, best quality, highres:1), peaceful beautiful sea scene",
"../../data/images/seaside4.jpeg",
],
]
with gr.Row():
gr.Examples(
examples=i2v_examples_256,
inputs=[prompt, image],
outputs=[out],
fn=hf_online_t2v_inference,
cache_examples=False,
)
img_edge_ratio.change(
fn=limit_shape,
inputs=[image, w, h, img_edge_ratio],
outputs=[img_edge_ratio_infact, out_w, out_h],
)
video_length.change(
fn=limit_length, inputs=[video_length], outputs=[video_length]
)
btn1.click(
fn=hf_online_t2v_inference,
inputs=[
prompt,
image,
seed,
fps,
w,
h,
video_length,
img_edge_ratio_infact,
],
outputs=out,
)
with gr.Tab("Video to Video"):
if ignore_video2video:
gr.Markdown(
(
"Due to GPU limit, MuseVDemo now only support Text2Video. If you want to try Video2Video, please run it locally. \n"
"We are trying to support video2video in the future. Thanks for your understanding."
)
)
else:
with gr.Row():
with gr.Column():
prompt = gr.Textbox(label="Prompt")
gr.Markdown(
(
"pose of VisionCondImage should be same as of the first frame of the video. "
"its better generate target first frame whose pose is same as of first frame of the video with text2image tool, sch as MJ, SDXL."
)
)
image = gr.Image(label="VisionCondImage")
video = gr.Video(label="ReferVideo")
# radio = gr.inputs.Radio(, label="Select an option")
# ctr_button = gr.inputs.Button(label="Add ControlNet List")
# output_text = gr.outputs.Textbox()
processor = gr.Textbox(
label=f"Control Condition. gradio code now only support dwpose_body_hand, use command can support multi of {control_options}",
value="dwpose_body_hand",
)
gr.Markdown("seed=-1 means that seeds are different in every run")
seed = gr.Number(
label="Seed (seed=-1 means that the seeds run each time are different)",
value=-1,
)
video_length = gr.Number(label="Video Length", value=12)
fps = gr.Number(label="Generate Video FPS", value=6)
gr.Markdown(
(
"If W&H is -1, then use the Reference Image's Size. Size of target video is $(W, H)*img\_edge\_ratio$. \n"
"The shorter the image size, the larger the motion amplitude, and the lower video quality.\n"
"The longer the W&H, the smaller the motion amplitude, and the higher video quality.\n"
"Due to the GPU VRAM limits, the W&H need smaller than 2000px"
)
)
with gr.Row():
w = gr.Number(label="Width", value=-1)
h = gr.Number(label="Height", value=-1)
img_edge_ratio = gr.Number(label="img_edge_ratio", value=1.0)
with gr.Row():
out_w = gr.Number(label="Width", value=0, interactive=False)
out_h = gr.Number(label="Height", value=0, interactive=False)
img_edge_ratio_infact = gr.Number(
label="img_edge_ratio in fact",
value=1.0,
interactive=False,
)
btn2 = gr.Button("Generate")
out1 = gr.Video()
v2v_examples_256 = [
[
"(masterpiece, best quality, highres:1), harley quinn is dancing, animation, by joshua klein",
"../../data/demo/cyber_girl.png",
"../../data/demo/video1.mp4",
],
]
with gr.Row():
gr.Examples(
examples=v2v_examples_256,
inputs=[prompt, image, video],
outputs=[out],
fn=hg_online_v2v_inference,
cache_examples=False,
)
img_edge_ratio.change(
fn=limit_shape,
inputs=[image, w, h, img_edge_ratio],
outputs=[img_edge_ratio_infact, out_w, out_h],
)
video_length.change(
fn=limit_length, inputs=[video_length], outputs=[video_length]
)
btn2.click(
fn=hg_online_v2v_inference,
inputs=[
prompt,
image,
video,
processor,
seed,
fps,
w,
h,
video_length,
img_edge_ratio_infact,
],
outputs=out1,
)
# Set the IP and port
ip_address = "0.0.0.0" # Replace with your desired IP address
port_number = 7860 # Replace with your desired port number
demo.queue().launch(
share=True, debug=True, server_name=ip_address, server_port=port_number
)
|