Spaces:
Running
on
Zero
Running
on
Zero
File size: 57,105 Bytes
2a00960 9be95d4 380884d 2a00960 2cde909 2a00960 947238d 2a00960 6cad0b7 2a00960 6cad0b7 2a00960 121fe52 2a00960 dd9f83d 3bddcc3 8ac9ea0 dd9f83d 2a00960 fd61736 2a00960 2cde909 2a00960 fd61736 2a00960 121fe52 2a00960 71d8df5 2a00960 71d8df5 2a00960 a7ce971 2a00960 85ff3b1 2a00960 71d8df5 2a00960 2003a89 2a00960 121fe52 2a00960 2003a89 121fe52 2a00960 fd61736 2a00960 71d8df5 2a00960 71d8df5 2a00960 3c9306c 2a00960 85ff3b1 2a00960 71d8df5 2a00960 27db4aa 2a00960 34fed27 27db4aa 34fed27 2a00960 71d8df5 2a00960 71d8df5 2a00960 71d8df5 2a00960 71d8df5 2a00960 71d8df5 2a00960 71d8df5 2a00960 71d8df5 2a00960 6cad0b7 2a00960 6cad0b7 2a00960 947238d |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 |
# -*- coding: utf-8 -*-
# Copyright (c) Alibaba, Inc. and its affiliates.
import os
import shlex
import subprocess
subprocess.run(shlex.split('pip install flash-attn --no-build-isolation'), env=os.environ | {'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"})
import sys
import csv
csv.field_size_limit(sys.maxsize)
import argparse
import base64
import copy
import glob
import io
import os
import random
import re
import string
import threading
import spaces
import cv2
import gradio as gr
import numpy as np
import torch
import transformers
from diffusers import CogVideoXImageToVideoPipeline
from diffusers.utils import export_to_video
from gradio_imageslider import ImageSlider
from PIL import Image
from transformers import AutoModel, AutoTokenizer
from scepter.modules.utils.config import Config
from scepter.modules.utils.directory import get_md5
from scepter.modules.utils.file_system import FS
from scepter.studio.utils.env import init_env
from infer import ACEInference
from example import get_examples
from utils import load_image
refresh_sty = '\U0001f504' # 🔄
clear_sty = '\U0001f5d1' # 🗑️
upload_sty = '\U0001f5bc' # 🖼️
sync_sty = '\U0001f4be' # 💾
chat_sty = '\U0001F4AC' # 💬
video_sty = '\U0001f3a5' # 🎥
lock = threading.Lock()
class ChatBotUI(object):
def __init__(self,
cfg,
root_work_dir='./'):
cfg.WORK_DIR = os.path.join(root_work_dir, cfg.WORK_DIR)
if not FS.exists(cfg.WORK_DIR):
FS.make_dir(cfg.WORK_DIR)
cfg = init_env(cfg)
self.cache_dir = cfg.WORK_DIR
self.chatbot_examples = get_examples(self.cache_dir)
self.model_cfg_dir = cfg.MODEL.EDIT_MODEL.MODEL_CFG_DIR
self.model_yamls = glob.glob(os.path.join(self.model_cfg_dir,
'*.yaml'))
self.model_choices = dict()
for i in self.model_yamls:
model_name = '.'.join(i.split('/')[-1].split('.')[:-1])
self.model_choices[model_name] = i
print('Models: ', self.model_choices)
self.model_name = cfg.MODEL.EDIT_MODEL.DEFAULT
assert self.model_name in self.model_choices
model_cfg = Config(load=True,
cfg_file=self.model_choices[self.model_name])
self.pipe = ACEInference()
self.pipe.init_from_cfg(model_cfg)
self.max_msgs = 20
self.enable_i2v = cfg.get('ENABLE_I2V', False)
if self.enable_i2v:
self.i2v_model_dir = cfg.MODEL.I2V.MODEL_DIR
self.i2v_model_name = cfg.MODEL.I2V.MODEL_NAME
if self.i2v_model_name == 'CogVideoX-5b-I2V':
with FS.get_dir_to_local_dir(self.i2v_model_dir) as local_dir:
self.i2v_pipe = CogVideoXImageToVideoPipeline.from_pretrained(
local_dir, torch_dtype=torch.bfloat16).cuda()
else:
raise NotImplementedError
with FS.get_dir_to_local_dir(
cfg.MODEL.CAPTIONER.MODEL_DIR) as local_dir:
self.captioner = AutoModel.from_pretrained(
local_dir,
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
use_flash_attn=True,
trust_remote_code=True).eval().cuda()
self.llm_tokenizer = AutoTokenizer.from_pretrained(
local_dir, trust_remote_code=True, use_fast=False)
self.llm_generation_config = dict(max_new_tokens=1024,
do_sample=True)
self.llm_prompt = cfg.LLM.PROMPT
self.llm_max_num = 2
with FS.get_dir_to_local_dir(
cfg.MODEL.ENHANCER.MODEL_DIR) as local_dir:
self.enhancer = transformers.pipeline(
'text-generation',
model=local_dir,
model_kwargs={'torch_dtype': torch.bfloat16},
device_map='auto',
)
sys_prompt = """You are part of a team of bots that creates videos. You work with an assistant bot that will draw anything you say in square brackets.
For example , outputting " a beautiful morning in the woods with the sun peaking through the trees " will trigger your partner bot to output an video of a forest morning , as described. You will be prompted by people looking to create detailed , amazing videos. The way to accomplish this is to take their short prompts and make them extremely detailed and descriptive.
There are a few rules to follow:
You will only ever output a single video description per user request.
When modifications are requested , you should not simply make the description longer . You should refactor the entire description to integrate the suggestions.
Other times the user will not want modifications , but instead want a new image . In this case , you should ignore your previous conversation with the user.
Video descriptions must have the same num of words as examples below. Extra words will be ignored.
"""
self.enhance_ctx = [
{
'role': 'system',
'content': sys_prompt
},
{
'role':
'user',
'content':
'Create an imaginative video descriptive caption or modify an earlier caption for the user input : "a girl is on the beach"',
},
{
'role':
'assistant',
'content':
"A radiant woman stands on a deserted beach, arms outstretched, wearing a beige trench coat, white blouse, light blue jeans, and chic boots, against a backdrop of soft sky and sea. Moments later, she is seen mid-twirl, arms exuberant, with the lighting suggesting dawn or dusk. Then, she runs along the beach, her attire complemented by an off-white scarf and black ankle boots, the tranquil sea behind her. Finally, she holds a paper airplane, her pose reflecting joy and freedom, with the ocean's gentle waves and the sky's soft pastel hues enhancing the serene ambiance.",
},
{
'role':
'user',
'content':
'Create an imaginative video descriptive caption or modify an earlier caption for the user input : "A man jogging on a football field"',
},
{
'role':
'assistant',
'content':
"A determined man in athletic attire, including a blue long-sleeve shirt, black shorts, and blue socks, jogs around a snow-covered soccer field, showcasing his solitary exercise in a quiet, overcast setting. His long dreadlocks, focused expression, and the serene winter backdrop highlight his dedication to fitness. As he moves, his attire, consisting of a blue sports sweatshirt, black athletic pants, gloves, and sneakers, grips the snowy ground. He is seen running past a chain-link fence enclosing the playground area, with a basketball hoop and children's slide, suggesting a moment of solitary exercise amidst the empty field.",
},
{
'role':
'user',
'content':
'Create an imaginative video descriptive caption or modify an earlier caption for the user input : " A woman is dancing, HD footage, close-up"',
},
{
'role':
'assistant',
'content':
'A young woman with her hair in an updo and wearing a teal hoodie stands against a light backdrop, initially looking over her shoulder with a contemplative expression. She then confidently makes a subtle dance move, suggesting rhythm and movement. Next, she appears poised and focused, looking directly at the camera. Her expression shifts to one of introspection as she gazes downward slightly. Finally, she dances with confidence, her left hand over her heart, symbolizing a poignant moment, all while dressed in the same teal hoodie against a plain, light-colored background.',
},
]
def create_ui(self):
css = '.chatbot.prose.md {opacity: 1.0 !important} #chatbot {opacity: 1.0 !important}'
with gr.Blocks(css=css,
title='Chatbot',
head='Chatbot',
analytics_enabled=False):
self.history = gr.State(value=[])
self.images = gr.State(value={})
self.history_result = gr.State(value={})
self.retry_msg = gr.State(value='')
with gr.Group():
with gr.Row(equal_height=True):
with gr.Column(visible=True) as self.chat_page:
self.chatbot = gr.Chatbot(
height=600,
value=[],
bubble_full_width=False,
show_copy_button=True,
container=False,
placeholder='<strong>Chat Box</strong>')
with gr.Row():
self.clear_btn = gr.Button(clear_sty +
' Clear Chat',
size='sm')
with gr.Column(visible=False) as self.editor_page:
with gr.Tabs():
with gr.Tab(id='ImageUploader',
label='Image Uploader',
visible=True) as self.upload_tab:
self.image_uploader = gr.Image(
height=550,
interactive=True,
type='pil',
image_mode='RGB',
sources='upload',
elem_id='image_uploader',
format='png')
with gr.Row():
self.sub_btn_1 = gr.Button(
value='Submit',
elem_id='upload_submit')
self.ext_btn_1 = gr.Button(value='Exit')
with gr.Tab(id='ImageEditor',
label='Image Editor',
visible=False) as self.edit_tab:
self.mask_type = gr.Dropdown(
label='Mask Type',
choices=[
'Background', 'Composite',
'Outpainting'
],
value='Background')
self.mask_type_info = gr.HTML(
value=
"<div style='background-color: white; padding-left: 15px; color: grey;'>Background mode will not erase the visual content in the mask area</div>"
)
with gr.Accordion(
label='Outpainting Setting',
open=True,
visible=False) as self.outpaint_tab:
with gr.Row(variant='panel'):
self.top_ext = gr.Slider(
show_label=True,
label='Top Extend Ratio',
minimum=0.0,
maximum=2.0,
step=0.1,
value=0.25)
self.bottom_ext = gr.Slider(
show_label=True,
label='Bottom Extend Ratio',
minimum=0.0,
maximum=2.0,
step=0.1,
value=0.25)
with gr.Row(variant='panel'):
self.left_ext = gr.Slider(
show_label=True,
label='Left Extend Ratio',
minimum=0.0,
maximum=2.0,
step=0.1,
value=0.25)
self.right_ext = gr.Slider(
show_label=True,
label='Right Extend Ratio',
minimum=0.0,
maximum=2.0,
step=0.1,
value=0.25)
with gr.Row(variant='panel'):
self.img_pad_btn = gr.Button(
value='Pad Image')
self.image_editor = gr.ImageMask(
value=None,
sources=[],
layers=False,
label='Edit Image',
elem_id='image_editor',
format='png')
with gr.Row():
self.sub_btn_2 = gr.Button(
value='Submit', elem_id='edit_submit')
self.ext_btn_2 = gr.Button(value='Exit')
with gr.Tab(id='ImageViewer',
label='Image Viewer',
visible=False) as self.image_view_tab:
self.image_viewer = ImageSlider(
label='Image',
type='pil',
show_download_button=True,
elem_id='image_viewer')
self.ext_btn_3 = gr.Button(value='Exit')
with gr.Tab(id='VideoViewer',
label='Video Viewer',
visible=False) as self.video_view_tab:
self.video_viewer = gr.Video(
label='Video',
interactive=False,
sources=[],
format='mp4',
show_download_button=True,
elem_id='video_viewer',
loop=True,
autoplay=True)
self.ext_btn_4 = gr.Button(value='Exit')
with gr.Accordion(label='Setting', open=False):
with gr.Row():
self.model_name_dd = gr.Dropdown(
choices=self.model_choices,
value=self.model_name,
label='Model Version')
with gr.Row():
self.negative_prompt = gr.Textbox(
value='',
placeholder=
'Negative prompt used for Classifier-Free Guidance',
label='Negative Prompt',
container=False)
with gr.Row():
with gr.Column(scale=8, min_width=500):
with gr.Row():
self.step = gr.Slider(minimum=1,
maximum=1000,
value=20,
label='Sample Step')
self.cfg_scale = gr.Slider(
minimum=1.0,
maximum=20.0,
value=4.5,
label='Guidance Scale')
self.rescale = gr.Slider(minimum=0.0,
maximum=1.0,
value=0.5,
label='Rescale')
self.seed = gr.Slider(minimum=-1,
maximum=10000000,
value=-1,
label='Seed')
self.output_height = gr.Slider(
minimum=256,
maximum=1024,
value=512,
label='Output Height')
self.output_width = gr.Slider(
minimum=256,
maximum=1024,
value=512,
label='Output Width')
with gr.Column(scale=1, min_width=50):
self.use_history = gr.Checkbox(value=False,
label='Use History')
self.video_auto = gr.Checkbox(
value=False,
label='Auto Gen Video',
visible=self.enable_i2v)
with gr.Row(variant='panel',
equal_height=True,
visible=self.enable_i2v):
self.video_fps = gr.Slider(minimum=1,
maximum=16,
value=8,
label='Video FPS',
visible=True)
self.video_frames = gr.Slider(minimum=8,
maximum=49,
value=49,
label='Video Frame Num',
visible=True)
self.video_step = gr.Slider(minimum=1,
maximum=1000,
value=50,
label='Video Sample Step',
visible=True)
self.video_cfg_scale = gr.Slider(
minimum=1.0,
maximum=20.0,
value=6.0,
label='Video Guidance Scale',
visible=True)
self.video_seed = gr.Slider(minimum=-1,
maximum=10000000,
value=-1,
label='Video Seed',
visible=True)
with gr.Row():
inst = """
**Instruction**:
1. Click 'Upload' button to upload one or more images as input images.
2. Enter '@' in the text box will exhibit all images in the gallery.
3. Select the image you wish to edit from the gallery, and its Image ID will be displayed in the text box.
4. Compose the editing instruction for the selected image, incorporating image id '@xxxxxx' into your instruction.
For example, you might say, "Change the girl's skirt in @123456 to blue." The '@xxxxx' token will facilitate the identification of the specific image, and will be automatically replaced by a special token '{image}' in the instruction.
5. Once your instructions are prepared, please click the "Chat" button to view the edited result in the chat window.
6. **Important** To render text on an image, please ensure to include a space between each letter. For instance, "add text 'g i r l' on the mask area of @xxxxx".
"""
gr.Markdown(value=inst)
with gr.Row(variant='panel',
equal_height=True,
show_progress=False):
with gr.Column(scale=1, min_width=100):
self.upload_btn = gr.Button(value=upload_sty +
' Upload',
variant='secondary')
with gr.Column(scale=5, min_width=500):
self.text = gr.Textbox(
placeholder='Input "@" find history of image',
label='Instruction',
container=False)
with gr.Column(scale=1, min_width=100):
self.chat_btn = gr.Button(value=chat_sty + ' Chat',
variant='primary')
with gr.Column(scale=1, min_width=100):
self.retry_btn = gr.Button(value=refresh_sty +
' Retry',
variant='secondary')
with gr.Column(scale=(1 if self.enable_i2v else 0),
min_width=0):
self.video_gen_btn = gr.Button(value=video_sty +
' Gen Video',
variant='secondary',
visible=self.enable_i2v)
with gr.Column(scale=(1 if self.enable_i2v else 0),
min_width=0):
self.extend_prompt = gr.Checkbox(
value=True,
label='Extend Prompt',
visible=self.enable_i2v)
with gr.Row():
self.gallery = gr.Gallery(visible=False,
label='History',
columns=10,
allow_preview=False,
interactive=False)
self.eg = gr.Column(visible=True)
def set_callbacks(self, *args, **kwargs):
########################################
@spaces.GPU(duration=60)
def change_model(model_name):
if model_name not in self.model_choices:
gr.Info('The provided model name is not a valid choice!')
return model_name, gr.update(), gr.update()
if model_name != self.model_name:
lock.acquire()
del self.pipe
torch.cuda.empty_cache()
model_cfg = Config(load=True,
cfg_file=self.model_choices[model_name])
self.pipe = ACEInference()
self.pipe.init_from_cfg(model_cfg)
self.model_name = model_name
lock.release()
return model_name, gr.update(), gr.update()
self.model_name_dd.change(
change_model,
inputs=[self.model_name_dd],
outputs=[self.model_name_dd, self.chatbot, self.text])
########################################
def generate_gallery(text, images):
if text.endswith(' '):
return gr.update(), gr.update(visible=False)
elif text.endswith('@'):
gallery_info = []
for image_id, image_meta in images.items():
thumbnail_path = image_meta['thumbnail']
gallery_info.append((thumbnail_path, image_id))
return gr.update(), gr.update(visible=True, value=gallery_info)
else:
gallery_info = []
match = re.search('@([^@ ]+)$', text)
if match:
prefix = match.group(1)
for image_id, image_meta in images.items():
if not image_id.startswith(prefix):
continue
thumbnail_path = image_meta['thumbnail']
gallery_info.append((thumbnail_path, image_id))
if len(gallery_info) > 0:
return gr.update(), gr.update(visible=True,
value=gallery_info)
else:
return gr.update(), gr.update(visible=False)
else:
return gr.update(), gr.update(visible=False)
self.text.input(generate_gallery,
inputs=[self.text, self.images],
outputs=[self.text, self.gallery],
show_progress='hidden')
########################################
def select_image(text, evt: gr.SelectData):
image_id = evt.value['caption']
text = '@'.join(text.split('@')[:-1]) + f'@{image_id} '
return gr.update(value=text), gr.update(visible=False, value=None)
self.gallery.select(select_image,
inputs=self.text,
outputs=[self.text, self.gallery])
########################################
@spaces.GPU(duration=120)
def generate_video(message,
extend_prompt,
history,
images,
num_steps,
num_frames,
cfg_scale,
fps,
seed,
progress=gr.Progress(track_tqdm=True)):
generator = torch.Generator(device='cuda').manual_seed(seed)
img_ids = re.findall('@(.*?)[ ,;.?$]', message)
if len(img_ids) == 0:
history.append((
message,
'Sorry, no images were found in the prompt to be used as the first frame of the video.'
))
while len(history) >= self.max_msgs:
history.pop(0)
return history, self.get_history(
history), gr.update(), gr.update(visible=False)
img_id = img_ids[0]
prompt = re.sub(f'@{img_id}\s+', '', message)
if extend_prompt:
messages = copy.deepcopy(self.enhance_ctx)
messages.append({
'role':
'user',
'content':
f'Create an imaginative video descriptive caption or modify an earlier caption in ENGLISH for the user input: "{prompt}"',
})
lock.acquire()
outputs = self.enhancer(
messages,
max_new_tokens=200,
)
prompt = outputs[0]['generated_text'][-1]['content']
print(prompt)
lock.release()
img_meta = images[img_id]
img_path = img_meta['image']
image = Image.open(img_path).convert('RGB')
lock.acquire()
video = self.i2v_pipe(
prompt=prompt,
image=image,
num_videos_per_prompt=1,
num_inference_steps=num_steps,
num_frames=num_frames,
guidance_scale=cfg_scale,
generator=generator,
).frames[0]
lock.release()
out_video_path = export_to_video(video, fps=fps)
history.append((
f"Based on first frame @{img_id} and description '{prompt}', generate a video",
'This is generated video:'))
history.append((None, out_video_path))
while len(history) >= self.max_msgs:
history.pop(0)
return history, self.get_history(history), gr.update(
value=''), gr.update(visible=False)
self.video_gen_btn.click(
generate_video,
inputs=[
self.text, self.extend_prompt, self.history, self.images,
self.video_step, self.video_frames, self.video_cfg_scale,
self.video_fps, self.video_seed
],
outputs=[self.history, self.chatbot, self.text, self.gallery])
########################################
@spaces.GPU(duration=60)
def run_chat(message,
extend_prompt,
history,
images,
use_history,
history_result,
negative_prompt,
cfg_scale,
rescale,
step,
seed,
output_h,
output_w,
video_auto,
video_steps,
video_frames,
video_cfg_scale,
video_fps,
video_seed,
progress=gr.Progress(track_tqdm=True)):
retry_msg = message
gen_id = get_md5(message)[:12]
save_path = os.path.join(self.cache_dir, f'{gen_id}.png')
img_ids = re.findall('@(.*?)[ ,;.?$]', message)
history_io = None
new_message = message
if len(img_ids) > 0:
edit_image, edit_image_mask, edit_task = [], [], []
for i, img_id in enumerate(img_ids):
if img_id not in images:
gr.Info(
f'The input image ID {img_id} is not exist... Skip loading image.'
)
continue
placeholder = '{image}' if i == 0 else '{' + f'image{i}' + '}'
new_message = re.sub(f'@{img_id}', placeholder,
new_message)
img_meta = images[img_id]
img_path = img_meta['image']
img_mask = img_meta['mask']
img_mask_type = img_meta['mask_type']
if img_mask_type is not None and img_mask_type == 'Composite':
task = 'inpainting'
else:
task = ''
edit_image.append(Image.open(img_path).convert('RGB'))
edit_image_mask.append(
Image.open(img_mask).
convert('L') if img_mask is not None else None)
edit_task.append(task)
if use_history and (img_id in history_result):
history_io = history_result[img_id]
buffered = io.BytesIO()
edit_image[0].save(buffered, format='JPEG')
img_b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
img_str = f'<img src="data:image/jpg;base64,{img_b64}" style="pointer-events: none;">'
pre_info = f'Received one or more images, so image editing is conducted.\n The first input image @{img_ids[0]} is:\n {img_str}'
else:
pre_info = 'No image ids were found in the provided text prompt, so text-guided image generation is conducted. \n'
edit_image = None
edit_image_mask = None
edit_task = ''
print(new_message)
imgs = self.pipe(
image=edit_image,
mask=edit_image_mask,
task=edit_task,
prompt=[new_message] *
len(edit_image) if edit_image is not None else [new_message],
negative_prompt=[negative_prompt] * len(edit_image)
if edit_image is not None else [negative_prompt],
history_io=history_io,
output_height=output_h,
output_width=output_w,
sampler='ddim',
sample_steps=step,
guide_scale=cfg_scale,
guide_rescale=rescale,
seed=seed,
)
img = imgs[0]
img.save(save_path, format='PNG')
if history_io:
history_io_new = copy.deepcopy(history_io)
history_io_new['image'] += edit_image[:1]
history_io_new['mask'] += edit_image_mask[:1]
history_io_new['task'] += edit_task[:1]
history_io_new['prompt'] += [new_message]
history_io_new['image'] = history_io_new['image'][-5:]
history_io_new['mask'] = history_io_new['mask'][-5:]
history_io_new['task'] = history_io_new['task'][-5:]
history_io_new['prompt'] = history_io_new['prompt'][-5:]
history_result[gen_id] = history_io_new
elif edit_image is not None and len(edit_image) > 0:
history_io_new = {
'image': edit_image[:1],
'mask': edit_image_mask[:1],
'task': edit_task[:1],
'prompt': [new_message]
}
history_result[gen_id] = history_io_new
w, h = img.size
if w > h:
tb_w = 128
tb_h = int(h * tb_w / w)
else:
tb_h = 128
tb_w = int(w * tb_h / h)
thumbnail_path = os.path.join(self.cache_dir,
f'{gen_id}_thumbnail.jpg')
thumbnail = img.resize((tb_w, tb_h))
thumbnail.save(thumbnail_path, format='JPEG')
images[gen_id] = {
'image': save_path,
'mask': None,
'mask_type': None,
'thumbnail': thumbnail_path
}
buffered = io.BytesIO()
img.convert('RGB').save(buffered, format='JPEG')
img_b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
img_str = f'<img src="data:image/jpg;base64,{img_b64}" style="pointer-events: none;">'
history.append(
(message,
f'{pre_info} The generated image @{gen_id} is:\n {img_str}'))
if video_auto:
if video_seed is None or video_seed == -1:
video_seed = random.randint(0, 10000000)
lock.acquire()
generator = torch.Generator(
device='cuda').manual_seed(video_seed)
pixel_values = load_image(img.convert('RGB'),
max_num=self.llm_max_num).to(
torch.bfloat16).cuda()
prompt = self.captioner.chat(self.llm_tokenizer, pixel_values,
self.llm_prompt,
self.llm_generation_config)
print(prompt)
lock.release()
if extend_prompt:
messages = copy.deepcopy(self.enhance_ctx)
messages.append({
'role':
'user',
'content':
f'Create an imaginative video descriptive caption or modify an earlier caption in ENGLISH for the user input: "{prompt}"',
})
lock.acquire()
outputs = self.enhancer(
messages,
max_new_tokens=200,
)
prompt = outputs[0]['generated_text'][-1]['content']
print(prompt)
lock.release()
lock.acquire()
video = self.i2v_pipe(
prompt=prompt,
image=img,
num_videos_per_prompt=1,
num_inference_steps=video_steps,
num_frames=video_frames,
guidance_scale=video_cfg_scale,
generator=generator,
).frames[0]
lock.release()
out_video_path = export_to_video(video, fps=video_fps)
history.append((
f"Based on first frame @{gen_id} and description '{prompt}', generate a video",
'This is generated video:'))
history.append((None, out_video_path))
while len(history) >= self.max_msgs:
history.pop(0)
return history, images, history_result, self.get_history(
history), gr.update(value=''), gr.update(
visible=False), retry_msg
chat_inputs = [
self.extend_prompt, self.history, self.images, self.use_history,
self.history_result, self.negative_prompt, self.cfg_scale,
self.rescale, self.step, self.seed, self.output_height,
self.output_width, self.video_auto, self.video_step,
self.video_frames, self.video_cfg_scale, self.video_fps,
self.video_seed
]
chat_outputs = [
self.history, self.images, self.history_result, self.chatbot,
self.text, self.gallery, self.retry_msg
]
self.chat_btn.click(run_chat,
inputs=[self.text] + chat_inputs,
outputs=chat_outputs)
self.text.submit(run_chat,
inputs=[self.text] + chat_inputs,
outputs=chat_outputs)
def retry_fn(*args):
return run_chat(*args)
self.retry_btn.click(retry_fn,
inputs=[self.retry_msg] + chat_inputs,
outputs=chat_outputs)
########################################
@spaces.GPU(duration=60)
def run_example(task, img, img_mask, ref1, prompt, seed):
edit_image, edit_image_mask, edit_task = [], [], []
if img is not None:
w, h = img.size
if w > 2048:
ratio = w / 2048.
w = 2048
h = int(h / ratio)
if h > 2048:
ratio = h / 2048.
h = 2048
w = int(w / ratio)
img = img.resize((w, h))
edit_image.append(img)
edit_image_mask.append(
img_mask if img_mask is not None else None)
edit_task.append(task)
if ref1 is not None:
edit_image.append(ref1)
edit_image_mask.append(None)
edit_task.append('')
buffered = io.BytesIO()
img.save(buffered, format='JPEG')
img_b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
img_str = f'<img src="data:image/jpg;base64,{img_b64}" style="pointer-events: none;">'
pre_info = f'Received one or more images, so image editing is conducted.\n The first input image is:\n {img_str}'
else:
pre_info = 'No image ids were found in the provided text prompt, so text-guided image generation is conducted. \n'
edit_image = None
edit_image_mask = None
edit_task = ''
img_num = len(edit_image) if edit_image is not None else 1
imgs = self.pipe(
image=edit_image,
mask=edit_image_mask,
task=edit_task,
prompt=[prompt] * img_num,
negative_prompt=[''] * img_num,
seed=seed,
)
img = imgs[0]
buffered = io.BytesIO()
img.convert('RGB').save(buffered, format='JPEG')
img_b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
img_str = f'<img src="data:image/jpg;base64,{img_b64}" style="pointer-events: none;">'
history = [(prompt,
f'{pre_info} The generated image is:\n {img_str}')]
return self.get_history(history), gr.update(value=''), gr.update(
visible=False), gr.update(value=-1)
with self.eg:
self.example_task = gr.Text(label='Task Name',
value='',
visible=False)
self.example_image = gr.Image(label='Edit Image',
type='pil',
image_mode='RGB',
visible=False)
self.example_mask = gr.Image(label='Edit Image Mask',
type='pil',
image_mode='L',
visible=False)
self.example_ref_im1 = gr.Image(label='Ref Image',
type='pil',
image_mode='RGB',
visible=False)
self.examples = gr.Examples(
fn=run_example,
examples=self.chatbot_examples,
inputs=[
self.example_task, self.example_image, self.example_mask,
self.example_ref_im1, self.text, self.seed
],
outputs=[self.chatbot, self.text, self.gallery, self.seed],
examples_per_page=4,
run_on_click=True)
########################################
def upload_image():
return (gr.update(visible=True,
scale=1), gr.update(visible=True, scale=1),
gr.update(visible=True), gr.update(visible=False),
gr.update(visible=False), gr.update(visible=False))
self.upload_btn.click(upload_image,
inputs=[],
outputs=[
self.chat_page, self.editor_page,
self.upload_tab, self.edit_tab,
self.image_view_tab, self.video_view_tab
])
########################################
def edit_image(evt: gr.SelectData):
if isinstance(evt.value, str):
img_b64s = re.findall(
'<img src="data:image/jpg;base64,(.*?)" style="pointer-events: none;">',
evt.value)
imgs = [
Image.open(io.BytesIO(base64.b64decode(copy.deepcopy(i))))
for i in img_b64s
]
if len(imgs) > 0:
if len(imgs) == 2:
view_img = copy.deepcopy(imgs)
edit_img = copy.deepcopy(imgs[-1])
else:
view_img = [
copy.deepcopy(imgs[-1]),
copy.deepcopy(imgs[-1])
]
edit_img = copy.deepcopy(imgs[-1])
return (gr.update(visible=True,
scale=1), gr.update(visible=True,
scale=1),
gr.update(visible=False), gr.update(visible=True),
gr.update(visible=True), gr.update(visible=False),
gr.update(value=edit_img),
gr.update(value=view_img), gr.update(value=None))
else:
return (gr.update(), gr.update(), gr.update(), gr.update(),
gr.update(), gr.update(), gr.update(), gr.update(),
gr.update())
elif isinstance(evt.value, dict) and evt.value.get(
'component', '') == 'video':
value = evt.value['value']['video']['path']
return (gr.update(visible=True,
scale=1), gr.update(visible=True, scale=1),
gr.update(visible=False), gr.update(visible=False),
gr.update(visible=False), gr.update(visible=True),
gr.update(), gr.update(), gr.update(value=value))
else:
return (gr.update(), gr.update(), gr.update(), gr.update(),
gr.update(), gr.update(), gr.update(), gr.update(),
gr.update())
self.chatbot.select(edit_image,
outputs=[
self.chat_page, self.editor_page,
self.upload_tab, self.edit_tab,
self.image_view_tab, self.video_view_tab,
self.image_editor, self.image_viewer,
self.video_viewer
])
self.image_viewer.change(lambda x: x,
inputs=self.image_viewer,
outputs=self.image_viewer)
########################################
def submit_upload_image(image, history, images):
history, images = self.add_uploaded_image_to_history(
image, history, images)
return gr.update(visible=False), gr.update(
visible=True), gr.update(
value=self.get_history(history)), history, images
self.sub_btn_1.click(
submit_upload_image,
inputs=[self.image_uploader, self.history, self.images],
outputs=[
self.editor_page, self.chat_page, self.chatbot, self.history,
self.images
])
########################################
def submit_edit_image(imagemask, mask_type, history, images):
history, images = self.add_edited_image_to_history(
imagemask, mask_type, history, images)
return gr.update(visible=False), gr.update(
visible=True), gr.update(
value=self.get_history(history)), history, images
self.sub_btn_2.click(submit_edit_image,
inputs=[
self.image_editor, self.mask_type,
self.history, self.images
],
outputs=[
self.editor_page, self.chat_page,
self.chatbot, self.history, self.images
])
########################################
def exit_edit():
return gr.update(visible=False), gr.update(visible=True, scale=3)
self.ext_btn_1.click(exit_edit,
outputs=[self.editor_page, self.chat_page])
self.ext_btn_2.click(exit_edit,
outputs=[self.editor_page, self.chat_page])
self.ext_btn_3.click(exit_edit,
outputs=[self.editor_page, self.chat_page])
self.ext_btn_4.click(exit_edit,
outputs=[self.editor_page, self.chat_page])
########################################
def update_mask_type_info(mask_type):
if mask_type == 'Background':
info = 'Background mode will not erase the visual content in the mask area'
visible = False
elif mask_type == 'Composite':
info = 'Composite mode will erase the visual content in the mask area'
visible = False
elif mask_type == 'Outpainting':
info = 'Outpaint mode is used for preparing input image for outpainting task'
visible = True
return (gr.update(
visible=True,
value=
f"<div style='background-color: white; padding-left: 15px; color: grey;'>{info}</div>"
), gr.update(visible=visible))
self.mask_type.change(update_mask_type_info,
inputs=self.mask_type,
outputs=[self.mask_type_info, self.outpaint_tab])
########################################
def extend_image(top_ratio, bottom_ratio, left_ratio, right_ratio,
image):
img = cv2.cvtColor(image['background'], cv2.COLOR_RGBA2RGB)
h, w = img.shape[:2]
new_h = int(h * (top_ratio + bottom_ratio + 1))
new_w = int(w * (left_ratio + right_ratio + 1))
start_h = int(h * top_ratio)
start_w = int(w * left_ratio)
new_img = np.zeros((new_h, new_w, 3), dtype=np.uint8)
new_mask = np.ones((new_h, new_w, 1), dtype=np.uint8) * 255
new_img[start_h:start_h + h, start_w:start_w + w, :] = img
new_mask[start_h:start_h + h, start_w:start_w + w] = 0
layer = np.concatenate([new_img, new_mask], axis=2)
value = {
'background': new_img,
'composite': new_img,
'layers': [layer]
}
return gr.update(value=value)
self.img_pad_btn.click(extend_image,
inputs=[
self.top_ext, self.bottom_ext,
self.left_ext, self.right_ext,
self.image_editor
],
outputs=self.image_editor)
########################################
def clear_chat(history, images, history_result):
history.clear()
images.clear()
history_result.clear()
return history, images, history_result, self.get_history(history)
self.clear_btn.click(
clear_chat,
inputs=[self.history, self.images, self.history_result],
outputs=[
self.history, self.images, self.history_result, self.chatbot
])
def get_history(self, history):
info = []
for item in history:
new_item = [None, None]
if isinstance(item[0], str) and item[0].endswith('.mp4'):
new_item[0] = gr.Video(item[0], format='mp4')
else:
new_item[0] = item[0]
if isinstance(item[1], str) and item[1].endswith('.mp4'):
new_item[1] = gr.Video(item[1], format='mp4')
else:
new_item[1] = item[1]
info.append(new_item)
return info
def generate_random_string(self, length=20):
letters_and_digits = string.ascii_letters + string.digits
random_string = ''.join(
random.choice(letters_and_digits) for i in range(length))
return random_string
def add_edited_image_to_history(self, image, mask_type, history, images):
if mask_type == 'Composite':
img = Image.fromarray(image['composite'])
else:
img = Image.fromarray(image['background'])
img_id = get_md5(self.generate_random_string())[:12]
save_path = os.path.join(self.cache_dir, f'{img_id}.png')
img.convert('RGB').save(save_path)
mask = image['layers'][0][:, :, 3]
mask = Image.fromarray(mask).convert('RGB')
mask_path = os.path.join(self.cache_dir, f'{img_id}_mask.png')
mask.save(mask_path)
w, h = img.size
if w > h:
tb_w = 128
tb_h = int(h * tb_w / w)
else:
tb_h = 128
tb_w = int(w * tb_h / h)
if mask_type == 'Background':
comp_mask = np.array(mask, dtype=np.uint8)
mask_alpha = (comp_mask[:, :, 0:1].astype(np.float32) *
0.6).astype(np.uint8)
comp_mask = np.concatenate([comp_mask, mask_alpha], axis=2)
thumbnail = Image.alpha_composite(
img.convert('RGBA'),
Image.fromarray(comp_mask).convert('RGBA')).convert('RGB')
else:
thumbnail = img.convert('RGB')
thumbnail_path = os.path.join(self.cache_dir,
f'{img_id}_thumbnail.jpg')
thumbnail = thumbnail.resize((tb_w, tb_h))
thumbnail.save(thumbnail_path, format='JPEG')
buffered = io.BytesIO()
img.convert('RGB').save(buffered, format='JPEG')
img_b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
img_str = f'<img src="data:image/jpg;base64,{img_b64}" style="pointer-events: none;">'
buffered = io.BytesIO()
mask.convert('RGB').save(buffered, format='JPEG')
mask_b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
mask_str = f'<img src="data:image/jpg;base64,{mask_b64}" style="pointer-events: none;">'
images[img_id] = {
'image': save_path,
'mask': mask_path,
'mask_type': mask_type,
'thumbnail': thumbnail_path
}
history.append((
None,
f'This is edited image and mask:\n {img_str} {mask_str} image ID is: {img_id}'
))
return history, images
def add_uploaded_image_to_history(self, img, history, images):
img_id = get_md5(self.generate_random_string())[:12]
save_path = os.path.join(self.cache_dir, f'{img_id}.png')
w, h = img.size
if w > 2048:
ratio = w / 2048.
w = 2048
h = int(h / ratio)
if h > 2048:
ratio = h / 2048.
h = 2048
w = int(w / ratio)
img = img.resize((w, h))
img.save(save_path)
w, h = img.size
if w > h:
tb_w = 128
tb_h = int(h * tb_w / w)
else:
tb_h = 128
tb_w = int(w * tb_h / h)
thumbnail_path = os.path.join(self.cache_dir,
f'{img_id}_thumbnail.jpg')
thumbnail = img.resize((tb_w, tb_h))
thumbnail.save(thumbnail_path, format='JPEG')
images[img_id] = {
'image': save_path,
'mask': None,
'mask_type': None,
'thumbnail': thumbnail_path
}
buffered = io.BytesIO()
img.convert('RGB').save(buffered, format='JPEG')
img_b64 = base64.b64encode(buffered.getvalue()).decode('utf-8')
img_str = f'<img src="data:image/jpg;base64,{img_b64}" style="pointer-events: none;">'
history.append(
(None,
f'This is uploaded image:\n {img_str} image ID is: {img_id}'))
return history, images
if __name__ == '__main__':
cfg = Config(cfg_file="config/chatbot_ui.yaml")
with gr.Blocks() as demo:
chatbot = ChatBotUI(cfg)
chatbot.create_ui()
chatbot.set_callbacks()
demo.launch()
|