forked from modelscope/facechain
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathapp.py
1581 lines (1341 loc) · 74.8 KB
/
app.py
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
# Copyright (c) Alibaba, Inc. and its affiliates.
import enum
import os
import json
import shutil
import slugify
import time
from concurrent.futures import ProcessPoolExecutor
import cv2
import gradio as gr
import numpy as np
import torch
from glob import glob
import platform
from facechain.utils import snapshot_download, check_ffmpeg, set_spawn_method, project_dir, join_worker_data_dir
from facechain.inference import preprocess_pose, GenPortrait
from facechain.inference_inpaint import GenPortrait_inpaint
from facechain.inference_talkinghead import SadTalker, text_to_speech_edge
from facechain.inference_tryon import GenPortrait_tryon
from facechain.train_text_to_image_lora import prepare_dataset, data_process_fn
from facechain.constants import neg_prompt as neg, pos_prompt_with_cloth, pos_prompt_with_style, \
pose_models, pose_examples, base_models, tts_speakers_map
training_done_count = 0
inference_done_count = 0
SDXL_BASE_MODEL_ID = 'AI-ModelScope/stable-diffusion-xl-base-1.0'
#character_model = 'AI-ModelScope/stable-diffusion-xl-base-1.0'
character_model = 'ly261666/cv_portrait_model'
BASE_MODEL_MAP = {
"leosamsMoonfilm_filmGrain20": "写实模型(Realistic sd_1.5 model)",
"MajicmixRealistic_v6": "\N{fire}写真模型(Photorealistic sd_1.5 model)",
"sdxl_1.0": "sdxl_1.0",
}
class UploadTarget(enum.Enum):
PERSONAL_PROFILE = 'Personal Profile'
LORA_LIaBRARY = 'LoRA Library'
# utils
def concatenate_images(images):
heights = [img.shape[0] for img in images]
max_width = sum([img.shape[1] for img in images])
concatenated_image = np.zeros((max(heights), max_width, 3), dtype=np.uint8)
x_offset = 0
for img in images:
concatenated_image[0:img.shape[0], x_offset:x_offset + img.shape[1], :] = img
x_offset += img.shape[1]
return concatenated_image
def select_function(evt: gr.SelectData):
name = evt.value[1] if isinstance(evt.value, (tuple, list)) else evt.value
matched = list(filter(lambda item: name == item['name'], styles))
style = matched[0]
return gr.Text.update(value=style['name'], visible=True)
def get_selected_image(state_image_list, evt: gr.SelectData):
return state_image_list[evt.index]
def update_prompt(style_model):
matched = list(filter(lambda item: style_model == item['name'], styles))
style = matched[0]
pos_prompt = generate_pos_prompt(style['name'], style['add_prompt_style'])
multiplier_style = style['multiplier_style']
multiplier_human = style['multiplier_human']
return gr.Textbox.update(value=pos_prompt), \
gr.Slider.update(value=multiplier_style), \
gr.Slider.update(value=multiplier_human)
def update_pose_model(pose_image, pose_model):
if pose_image is None:
return gr.Radio.update(value=pose_models[0]['name']), gr.Image.update(visible=False)
else:
if pose_model == 0:
pose_model = 1
pose_res_img = preprocess_pose(pose_image)
return gr.Radio.update(value=pose_models[pose_model]['name']), gr.Image.update(value=pose_res_img, visible=True)
def train_lora_fn(base_model_path=None, revision=None, sub_path=None, output_img_dir=None, work_dir=None, photo_num=0):
torch.cuda.empty_cache()
lora_r = 4
lora_alpha = 32
max_train_steps = min(photo_num * 200, 800)
if platform.system() == 'Windows':
if 'xl-base' in base_model_path:
command = [
'python', f'{project_dir}/facechain/train_text_to_image_lora_sdxl.py' if base_model_path is SDXL_BASE_MODEL_ID else f'{project_dir}/facechain/train_text_to_image_lora.py',
f'--pretrained_model_name_or_path={base_model_path}',
f'--revision={revision}',
f'--sub_path={sub_path}',
f'--output_dataset_name={output_img_dir}',
'--caption_column=text',
'--resolution=512',
'--random_flip',
'--train_batch_size=1',
'--num_train_epochs=200',
'--checkpointing_steps=5000',
'--learning_rate=1.5e-04',
'--lr_scheduler=cosine',
'--lr_warmup_steps=0',
'--seed=42',
f'--output_dir={work_dir}',
f'--lora_r={lora_r}',
f'--lora_alpha={lora_alpha}',
'--lora_text_encoder_r=32',
'--lora_text_encoder_alpha=32',
#'--use_swift',
#'--resume_from_checkpoint=fromfacecommon'
]
else:
command = [
'python', f'{project_dir}/facechain/train_text_to_image_lora_sdxl.py' if base_model_path is SDXL_BASE_MODEL_ID else f'{project_dir}/facechain/train_text_to_image_lora.py',
f'--pretrained_model_name_or_path={base_model_path}',
f'--revision={revision}',
f'--sub_path={sub_path}',
f'--output_dataset_name={output_img_dir}',
'--caption_column=text',
'--resolution=512',
'--random_flip',
'--train_batch_size=1',
'--num_train_epochs=200',
'--checkpointing_steps=5000',
'--learning_rate=1.5e-04',
'--lr_scheduler=cosine',
'--lr_warmup_steps=0',
'--seed=42',
f'--output_dir={work_dir}',
f'--lora_r={lora_r}',
f'--lora_alpha={lora_alpha}',
'--lora_text_encoder_r=32',
'--lora_text_encoder_alpha=32',
#'--use_swift',
'--resume_from_checkpoint=fromfacecommon'
]
import subprocess
try:
subprocess.run(command, check=True)
except subprocess.CalledProcessError as e:
print(f"Error executing the command: {e}")
raise gr.Error("训练失败 (Training failed)")
else:
print(f'** project dir: {project_dir}')
print(f'** params: >base_model_path:{base_model_path}, >revision:{revision}, >sub_path:{sub_path}, >output_img_dir:{output_img_dir}, >work_dir:{work_dir}, >lora_r:{lora_r}, >lora_alpha:{lora_alpha}')
train_script_path = f'{project_dir}/facechain/train_text_to_image_lora_sdxl.py' if base_model_path == SDXL_BASE_MODEL_ID else f'{project_dir}/facechain/train_text_to_image_lora.py'
if 'xl-base' in base_model_path:
res = os.system(
f'PYTHONPATH=. accelerate launch {train_script_path} '
f'--pretrained_model_name_or_path={base_model_path} '
f'--revision={revision} '
f'--sub_path={sub_path} '
f'--output_dataset_name={output_img_dir} '
f'--caption_column="text" '
f'--resolution=512 '
f'--random_flip '
f'--train_batch_size=1 '
f'--num_train_epochs=200 '
f'--checkpointing_steps=5000 '
f'--learning_rate=1.5e-04 '
f'--lr_scheduler="cosine" '
f'--lr_warmup_steps=0 '
f'--seed=42 '
f'--output_dir={work_dir} '
f'--lora_r={lora_r} '
f'--lora_alpha={lora_alpha} '
f'--lora_text_encoder_r=32 '
f'--lora_text_encoder_alpha=32 ')
#f'--use_swift '
#f'--resume_from_checkpoint="fromfacecommon"')
else:
res = os.system(
f'PYTHONPATH=. accelerate launch {train_script_path} '
f'--pretrained_model_name_or_path={base_model_path} '
f'--revision={revision} '
f'--sub_path={sub_path} '
f'--output_dataset_name={output_img_dir} '
f'--caption_column="text" '
f'--resolution=512 '
f'--random_flip '
f'--train_batch_size=1 '
f'--num_train_epochs=200 '
f'--checkpointing_steps=5000 '
f'--learning_rate=1.5e-04 '
f'--lr_scheduler="cosine" '
f'--lr_warmup_steps=0 '
f'--seed=42 '
f'--output_dir={work_dir} '
f'--lora_r={lora_r} '
f'--lora_alpha={lora_alpha} '
f'--lora_text_encoder_r=32 '
f'--lora_text_encoder_alpha=32 '
#f'--use_swift '
f'--resume_from_checkpoint="fromfacecommon"')
if res != 0:
raise gr.Error("训练失败 (Training failed)")
def generate_pos_prompt(style_model, prompt_cloth):
if style_model is not None:
matched = list(filter(lambda style: style_model == style['name'], styles))
if len(matched) == 0:
raise ValueError(f'styles not found: {style_model}')
matched = matched[0]
if matched['model_id'] is None:
pos_prompt = pos_prompt_with_cloth.format(prompt_cloth)
else:
pos_prompt = pos_prompt_with_style.format(matched['add_prompt_style'])
else:
pos_prompt = pos_prompt_with_cloth.format(prompt_cloth)
return pos_prompt
def launch_pipeline(uuid,
pos_prompt,
neg_prompt=None,
base_model_index=None,
user_model=None,
num_images=1,
lora_choice=None,
style_model=None,
multiplier_style=0.35,
multiplier_human=0.95,
pose_model=None,
pose_image=None,
sr_img_size=None,
cartoon_style_idx=None,
):
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
# Check base model
if base_model_index == None:
raise gr.Error('请选择基模型(Please select the base model)!')
set_spawn_method()
# Check character LoRA
tmp_character_model = base_models[base_model_index]['model_id']
if tmp_character_model != character_model:
tmp_character_model = 'AI-ModelScope/stable-diffusion-xl-base-1.0'
folder_path = join_worker_data_dir(uuid, tmp_character_model)
folder_list = []
if os.path.exists(folder_path):
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
file_lora_path_swift = f"{file_path}/swift"
if os.path.exists(file_lora_path) or os.path.exists(file_lora_path_swift):
folder_list.append(file)
if len(folder_list) == 0:
raise gr.Error('没有人物LoRA，请先训练(There is no character LoRA, please train first)!')
# Check output model
if user_model == None:
raise gr.Error('请选择人物LoRA(Please select the character LoRA)！')
# Check lora choice
if lora_choice == None:
raise gr.Error('请选择LoRA模型(Please select the LoRA model)!')
# Check style model
if style_model == None and lora_choice == 'preset':
raise gr.Error('请选择风格模型(Please select the style model)!')
base_model = base_models[base_model_index]['model_id']
revision = base_models[base_model_index]['revision']
sub_path = base_models[base_model_index]['sub_path']
before_queue_size = 0
before_done_count = inference_done_count
matched = list(filter(lambda item: style_model == item['name'], styles))
if len(matched) == 0:
raise ValueError(f'styles not found: {style_model}')
matched = matched[0]
style_model = matched['name']
if lora_choice == 'preset':
if matched['model_id'] is None:
style_model_path = None
else:
model_dir = snapshot_download(matched['model_id'], revision=matched['revision'])
style_model_path = os.path.join(model_dir, matched['bin_file'])
else:
print(f'uuid: {uuid}')
temp_lora_dir = join_worker_data_dir(uuid, 'temp_lora')
file_name = lora_choice
print(lora_choice.split('.')[-1], os.path.join(temp_lora_dir, file_name))
if lora_choice.split('.')[-1] != 'safetensors' or not os.path.exists(os.path.join(temp_lora_dir, file_name)):
raise ValueError(f'Invalid lora file: {lora_file.name}')
style_model_path = os.path.join(temp_lora_dir, file_name)
if pose_image is None or pose_model == 0:
pose_model_path = None
use_depth_control = False
pose_image = None
else:
model_dir = snapshot_download('damo/face_chain_control_model', revision='v1.0.1')
pose_model_path = os.path.join(model_dir, 'model_controlnet/control_v11p_sd15_openpose')
if pose_model == 1:
use_depth_control = True
else:
use_depth_control = False
print("-------user_model: ", user_model)
use_main_model = True
use_face_swap = True
use_post_process = True
use_stylization = False
instance_data_dir = join_worker_data_dir(uuid, 'training_data', tmp_character_model, user_model)
lora_model_path = join_worker_data_dir(uuid, tmp_character_model, user_model)
gen_portrait = GenPortrait(pose_model_path, pose_image, use_depth_control, pos_prompt, neg_prompt, style_model_path,
multiplier_style, multiplier_human, use_main_model,
use_face_swap, use_post_process,
use_stylization)
#gen_portrait(instance_data_dir, num_images, base_model, lora_model_path, sub_path, revision, sr_img_size, cartoon_style_idx )
num_images = min(6, num_images)
with ProcessPoolExecutor(max_workers=5) as executor:
future = executor.submit(gen_portrait, instance_data_dir,
num_images, base_model, lora_model_path, sub_path, revision, sr_img_size, cartoon_style_idx)
while not future.done():
is_processing = future.running()
if not is_processing:
cur_done_count = inference_done_count
to_wait = before_queue_size - (cur_done_count - before_done_count)
yield ["排队等待资源中, 前方还有{}个生成任务, 预计需要等待{}分钟...".format(to_wait, to_wait * 2.5),
None]
else:
yield ["生成中, 请耐心等待(Generating)...", None]
time.sleep(1)
outputs = future.result()
outputs_RGB = []
for out_tmp in outputs:
outputs_RGB.append(cv2.cvtColor(out_tmp, cv2.COLOR_BGR2RGB))
save_dir = join_worker_data_dir(uuid, 'inference_result', base_model, user_model)
if lora_choice == 'preset':
save_dir = os.path.join(save_dir, 'style_' + style_model)
else:
save_dir = os.path.join(save_dir, 'lora_' + os.path.basename(lora_choice).split('.')[0])
if not os.path.exists(save_dir):
os.makedirs(save_dir)
# use single to save outputs
if not os.path.exists(os.path.join(save_dir, 'single')):
os.makedirs(os.path.join(save_dir, 'single'))
for img in outputs:
# count the number of images in the folder
num = len(os.listdir(os.path.join(save_dir, 'single')))
cv2.imwrite(os.path.join(save_dir, 'single', str(num) + '.png'), img)
if len(outputs) > 0:
result = concatenate_images(outputs)
if not os.path.exists(os.path.join(save_dir, 'concat')):
os.makedirs(os.path.join(save_dir, 'concat'))
num = len(os.listdir(os.path.join(save_dir, 'concat')))
image_path = os.path.join(save_dir, 'concat', str(num) + '.png')
cv2.imwrite(image_path, result)
yield ["生成完毕(Generation done)!", outputs_RGB]
else:
yield ["生成失败, 请重试(Generation failed, please retry)!", outputs_RGB]
def launch_pipeline_inpaint(uuid,
base_model_index=None,
user_model_A=None,
user_model_B=None,
num_faces=1,
template_image=None):
before_queue_size = 0
before_done_count = inference_done_count
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
# Check base model
if base_model_index == None:
raise gr.Error('请选择基模型(Please select the base model)！')
# Check character LoRA
tmp_character_model = base_models[base_model_index]['model_id']
if tmp_character_model != character_model:
tmp_character_model = 'ly261666/cv_portrait_model'
folder_path = join_worker_data_dir(uuid, tmp_character_model)
folder_list = []
if os.path.exists(folder_path):
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
file_lora_path_swift = f"{file_path}/swift"
if os.path.exists(file_lora_path) or os.path.exists(file_lora_path_swift):
folder_list.append(file)
if len(folder_list) == 0:
raise gr.Error('没有人物LoRA，请先训练(There is no character LoRA, please train first)!')
# Check character LoRA
if num_faces == 1:
if user_model_A == None:
raise gr.Error('请至少选择一个人物LoRA(Please select at least one character LoRA)！')
else:
if user_model_A == None and user_model_B == None:
raise gr.Error('请至少选择一个人物LoRA(Please select at least one character LoRA)！')
if isinstance(template_image, str):
if len(template_image) == 0:
raise gr.Error('请选择一张模板(Please select 1 template)')
base_model = base_models[base_model_index]['model_id']
revision = base_models[base_model_index]['revision']
sub_path = base_models[base_model_index]['sub_path']
multiplier_style = 0.05
multiplier_human = 0.95
strength = 0.65
output_img_size = 512
model_dir = snapshot_download('ly261666/cv_wanx_style_model', revision='v1.0.3')
style_model_path = os.path.join(model_dir, 'zjz_mj_jiyi_small_addtxt_frommajicreal.safetensors')
pos_prompt = 'raw photo, masterpiece, chinese, simple background, high-class pure color background, solo, medium shot, high detail face, photorealistic, best quality, wearing T-shirt'
neg_prompt = 'nsfw, paintings, sketches, (worst quality:2), (low quality:2) ' \
'lowers, normal quality, ((monochrome)), ((grayscale)), logo, word, character'
if user_model_A == '不重绘该人物(Do not inpaint this character)':
user_model_A = None
if user_model_B == '不重绘该人物(Do not inpaint this character)':
user_model_B = None
if user_model_A is not None:
instance_data_dir_A = join_worker_data_dir(uuid, 'training_data', tmp_character_model, user_model_A)
lora_model_path_A = join_worker_data_dir(uuid, tmp_character_model, user_model_A)
else:
instance_data_dir_A = None
lora_model_path_A = None
if user_model_B is not None:
instance_data_dir_B = join_worker_data_dir(uuid, 'training_data', tmp_character_model, user_model_B)
lora_model_path_B = join_worker_data_dir(uuid, tmp_character_model, user_model_B)
else:
instance_data_dir_B = None
lora_model_path_B = None
in_path = template_image
out_path = 'inpaint_rst'
use_main_model = True
use_face_swap = True
use_post_process = True
use_stylization = False
gen_portrait = GenPortrait_inpaint(in_path, strength, num_faces,
pos_prompt, neg_prompt, style_model_path,
multiplier_style, multiplier_human, use_main_model,
use_face_swap, use_post_process,
use_stylization)
with ProcessPoolExecutor(max_workers=5) as executor:
future = executor.submit(gen_portrait, instance_data_dir_A, instance_data_dir_B, base_model,\
lora_model_path_A, lora_model_path_B, sub_path=sub_path, revision=revision)
while not future.done():
is_processing = future.running()
if not is_processing:
cur_done_count = inference_done_count
to_wait = before_queue_size - (cur_done_count - before_done_count)
yield ["排队等待资源中，前方还有{}个生成任务, 预计需要等待{}分钟...".format(to_wait, to_wait * 2.5),
None]
else:
yield ["生成中, 请耐心等待(Generating)...", None]
time.sleep(1)
outputs = future.result()
outputs_RGB = []
for out_tmp in outputs:
outputs_RGB.append(cv2.cvtColor(out_tmp, cv2.COLOR_BGR2RGB))
for i, out_tmp in enumerate(outputs):
cv2.imwrite('{}_{}.png'.format(out_path, i), out_tmp)
if len(outputs) > 0:
yield ["生成完毕(Generation done)！", outputs_RGB]
else:
yield ["生成失败，请重试(Generation failed, please retry)！", outputs_RGB]
def get_previous_image_result(uuid):
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
return "请登陆后使用! (Please login first)"
else:
uuid = 'qw'
save_dir_old = os.path.join('/tmp', uuid, 'inference_result')
image_results_old = glob(os.path.join(save_dir_old, '**/single/*.png'), recursive=True)
save_dir = os.path.join('.', uuid, 'inference_result')
image_results = glob(os.path.join(save_dir, '**/single/*.png'), recursive=True)
save_dir_new = join_worker_data_dir(uuid, 'inference_result')
image_results_new = glob(os.path.join(save_dir_new, '**/single/*.png'), recursive=True)
return image_results_old + image_results + image_results_new
def toggle_audio(choice):
if choice == "麦克风(microphone)":
return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), \
gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)
elif choice == "上传文件(upload)":
return gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), \
gr.update(visible=False), gr.update(visible=False), gr.update(visible=True)
else:
return gr.update(visible=True), gr.update(visible=True), gr.update(visible=True),\
gr.update(visible=True), gr.update(visible=False), gr.update(visible=False)
def launch_pipeline_talkinghead(uuid, source_image, audio_source, audio_tts, audio_microphone, audio_upload,
preprocess='crop', still_mode=True, use_enhancer=False, batch_size=1, size=256,
pose_style = 0, exp_scale=1.0):
if not check_ffmpeg():
raise gr.Error("请先安装ffmpeg，然后刷新网页（Please install ffmpeg, then restart the webpage）")
before_queue_size = 0
before_done_count = inference_done_count
if not source_image:
raise gr.Error('请选择一张源图片(Please select 1 source image)')
if audio_source == "语音合成(TTS)":
driven_audio = audio_tts
elif audio_source == "麦克风(microphone)":
driven_audio = audio_microphone
else:
driven_audio = audio_upload
if not driven_audio:
raise gr.Error('请提供一段wav、mp3音频(Please provide 1 wav or mp3 audio)')
user_directory = os.path.expanduser("~")
if not os.path.exists(os.path.join(user_directory, '.cache', 'modelscope', 'hub', 'wwd123', 'sadtalker')):
gr.Info("第一次初始化会比较耗时，请耐心等待(The first time initialization will take time, please wait)")
gen_video = SadTalker(uuid)
with ProcessPoolExecutor(max_workers=5) as executor:
future = executor.submit(gen_video, source_image, driven_audio, preprocess,
still_mode, use_enhancer, batch_size, size, pose_style, exp_scale)
while not future.done():
is_processing = future.running()
if not is_processing:
cur_done_count = inference_done_count
to_wait = before_queue_size - (cur_done_count - before_done_count)
yield ["排队等待资源中，前方还有{}个生成任务(Queueing, there are {} tasks ahead)".format(to_wait, to_wait),
None]
else:
yield ["生成中, 请耐心等待(Generating, please wait)...", None]
time.sleep(1)
output = future.result()
if output:
yield ["生成完毕(Generation done)！", output]
else:
yield ["生成失败，请重试(Generation failed, please retry)！", output]
def launch_pipeline_tryon(uuid,
user_model=None,
template_image=None,
background_prompt=None,
do_inpaint=1):
before_queue_size = 0
before_done_count = inference_done_count
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
# Check base model
base_model_index = 0
# Check character LoRA
folder_path = join_worker_data_dir(uuid, character_model)
folder_list = []
if os.path.exists(folder_path):
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
if os.path.exists(file_lora_path):
folder_list.append(file)
if len(folder_list) == 0:
raise gr.Error('没有人物LoRA，请先训练(There is no character LoRA, please train first)!')
# Check character LoRA
if 1:
if user_model == None:
raise gr.Error('请选择人物LoRA(Please select character LoRA)！')
if isinstance(template_image, str):
if len(template_image) == 0:
raise gr.Error('请选择一张模板(Please select 1 template)')
base_model = base_models[base_model_index]['model_id']
revision = base_models[base_model_index]['revision']
sub_path = base_models[base_model_index]['sub_path']
multiplier_style = 0.05
multiplier_human = 0.95
tmp_character_model = base_models[base_model_index]['model_id']
if tmp_character_model != character_model:
tmp_character_model = 'ly261666/cv_portrait_model'
model_dir = snapshot_download('ly261666/cv_wanx_style_model', revision='v1.0.3')
style_model_path = os.path.join(model_dir, 'zjz_mj_jiyi_small_addtxt_frommajicreal.safetensors')
pos_prompt = 'raw photo, masterpiece, ' + background_prompt + ', (completely nude:2), solo, medium shot, high detail face, photorealistic, best quality'
neg_prompt = 'nsfw, paintings, sketches, (worst quality:2), (low quality:2) ' \
'lowers, normal quality, ((monochrome)), ((grayscale)), logo, word, character'
if user_model == '不重绘该人物(Do not inpaint this character)':
user_model = None
if user_model is not None:
instance_data_dir = join_worker_data_dir(uuid, 'training_data', tmp_character_model, user_model)
lora_model_path = join_worker_data_dir(uuid, tmp_character_model, user_model)
else:
instance_data_dir = None
lora_model_path = None
in_path = template_image
out_path = 'inpaint_rst'
use_main_model = True
use_face_swap = True
use_post_process = True
use_stylization = False
gen_portrait = GenPortrait_tryon(in_path, 1.0,
pos_prompt, neg_prompt, style_model_path,
multiplier_style, multiplier_human, use_main_model,
use_face_swap, use_post_process,
use_stylization)
with ProcessPoolExecutor(max_workers=5) as executor:
future = executor.submit(gen_portrait, instance_data_dir, base_model, \
lora_model_path, sub_path=sub_path, revision=revision)
while not future.done():
is_processing = future.running()
if not is_processing:
cur_done_count = inference_done_count
to_wait = before_queue_size - (cur_done_count - before_done_count)
yield ["排队等待资源中，前方还有{}个生成任务, 预计需要等待{}分钟...".format(to_wait, to_wait * 2.5),
None]
else:
yield ["生成中, 请耐心等待(Generating)...", None]
time.sleep(1)
outputs = future.result()
if do_inpaint == 0:
cv2.imwrite('tmp_tryon.png', outputs[0])
pos_prompt = 'raw photo, masterpiece, chinese, simple background, high-class pure color background, solo, medium shot, high detail face, photorealistic, best quality, wearing T-shirt'
neg_prompt = 'nsfw, paintings, sketches, (worst quality:2), (low quality:2) ' \
'lowers, normal quality, ((monochrome)), ((grayscale)), logo, word, character'
gen_portrait = GenPortrait_inpaint('tmp_tryon.png', 0.65, 1,
pos_prompt, neg_prompt, style_model_path,
multiplier_style, multiplier_human, use_main_model,
use_face_swap, use_post_process,
use_stylization)
with ProcessPoolExecutor(max_workers=5) as executor:
future = executor.submit(gen_portrait, instance_data_dir, None, base_model, \
lora_model_path, None, sub_path=sub_path, revision=revision)
while not future.done():
is_processing = future.running()
if not is_processing:
cur_done_count = inference_done_count
to_wait = before_queue_size - (cur_done_count - before_done_count)
yield ["排队等待资源中，前方还有{}个生成任务, 预计需要等待{}分钟...".format(to_wait, to_wait * 2.5),
None]
else:
yield ["生成中, 请耐心等待(Generating)...", None]
time.sleep(1)
outputs = future.result()
outputs_RGB = []
for out_tmp in outputs:
outputs_RGB.append(cv2.cvtColor(out_tmp, cv2.COLOR_BGR2RGB))
for i, out_tmp in enumerate(outputs):
cv2.imwrite('{}_{}.png'.format(out_path, i), out_tmp)
if len(outputs) > 0:
yield ["生成完毕(Generation done)！", outputs_RGB]
else:
yield ["生成失败，请重试(Generation failed, please retry)！", outputs_RGB]
class Trainer:
def __init__(self):
pass
def run(
self,
uuid: str,
base_model_name: str,
instance_images: list,
output_model_name: str,
) -> str:
set_spawn_method()
# Check Cuda
if not torch.cuda.is_available():
raise gr.Error('CUDA不可用(CUDA not available)')
# Check Cuda Memory
if torch.cuda.is_available():
device = torch.device("cuda:0")
required_memory_bytes = 18 * (1024 ** 3) # 18GB
try:
tensor = torch.empty((required_memory_bytes // 4,), device = device) # create 18GB tensor to check the memory if enough
print("显存足够")
del tensor
except RuntimeError as e:
raise gr.Error("目前显存不足18GB，训练失败！")
# Check Instance Valid
if instance_images is None:
raise gr.Error('您需要上传训练图片(Please upload photos)!')
# Check output model name
if not output_model_name:
raise gr.Error('请指定人物lora的名称(Please specify the character LoRA name)！')
# Limit input Image
if len(instance_images) > 20:
raise gr.Error('请最多上传20张训练图片(20 images at most!)')
# Check UUID & Studio
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
return "请登陆后使用(Please login first)! "
else:
uuid = 'qw'
if base_model_name == SDXL_BASE_MODEL_ID:
print('** Setting base model to SDXL **')
base_model_path = SDXL_BASE_MODEL_ID
revision = 'v1.0.9'
sub_path = ''
else:
print('** Setting base model to SD1.5 **')
base_model_path = 'ly261666/cv_portrait_model'
revision = 'v2.0'
sub_path = "film/film"
output_model_name = slugify.slugify(output_model_name)
# mv user upload data to target dir
instance_data_dir = join_worker_data_dir(uuid, 'training_data', base_model_path, output_model_name)
print("--------uuid: ", uuid)
uuid_dir = join_worker_data_dir(uuid)
if not os.path.exists(uuid_dir):
os.makedirs(uuid_dir)
work_dir = join_worker_data_dir(uuid, base_model_path, output_model_name)
if os.path.exists(work_dir):
raise gr.Error("人物lora名称已存在。(This character lora name already exists.)")
print("----------work_dir: ", work_dir)
shutil.rmtree(work_dir, ignore_errors=True)
shutil.rmtree(instance_data_dir, ignore_errors=True)
prepare_dataset([img['name'] for img in instance_images], output_dataset_dir=instance_data_dir)
data_process_fn(instance_data_dir, True)
# train lora
print("instance_data_dir", instance_data_dir)
train_lora_fn(base_model_path=base_model_path,
revision=revision,
sub_path=sub_path,
output_img_dir=instance_data_dir,
work_dir=work_dir,
photo_num=len(instance_images))
message = '''<center><font size=4>训练已经完成！请切换至 [无限风格形象写真] 标签体验模型效果。</center>
<center><font size=4>(Training done, please switch to the Infinite Style Portrait tab to generate photos.)</center>'''
print(message)
return message
def flash_model_list(uuid, base_model_index, lora_choice:gr.Dropdown):
base_model_path = base_models[base_model_index]['model_id']
style_list = base_models[base_model_index]['style_list']
sub_styles=[]
for style in style_list:
matched = list(filter(lambda item: style == item['name'], styles))
sub_styles.append(matched[0])
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
folder_path = join_worker_data_dir(uuid, character_model)
folder_list = []
lora_save_path = join_worker_data_dir(uuid, 'temp_lora')
if not os.path.exists(lora_save_path):
lora_list = ['preset']
else:
lora_list = sorted(os.listdir(lora_save_path))
lora_list = ["preset"] + lora_list
if not os.path.exists(folder_path):
if lora_choice == 'preset':
return gr.Radio.update(choices=[], value = None), \
gr.Gallery.update(value=[(item["img"], item["name"]) for item in sub_styles], visible=True), \
gr.Text.update(value=style_list[0], visible=True), \
gr.Dropdown.update(choices=lora_list, visible=True), gr.File.update(visible=True)
else:
return gr.Radio.update(choices=[], value = None), \
gr.Gallery.update(visible=False), gr.Text.update(),\
gr.Dropdown.update(choices=lora_list, visible=True), gr.File.update(visible=True)
else:
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
file_lora_path_swift = f"{file_path}/unet"
if os.path.exists(file_lora_path) or os.path.exists(file_lora_path_swift):
folder_list.append(file)
if lora_choice == 'preset':
return gr.Radio.update(choices=folder_list, value = None), \
gr.Gallery.update(value=[(item["img"], item["name"]) for item in sub_styles], visible=True), \
gr.Text.update(value=style_list[0], visible=True), \
gr.Dropdown.update(choices=lora_list, visible=True), gr.File.update(visible=True)
else:
return gr.Radio.update(choices=folder_list, value = None), \
gr.Gallery.update(visible=False), gr.Text.update(), \
gr.Dropdown.update(choices=lora_list, visible=True), gr.File.update(visible=True)
def update_output_model(uuid):
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
folder_list = []
for idx, tmp_character_model in enumerate(['AI-ModelScope/stable-diffusion-xl-base-1.0', character_model]):
folder_path = join_worker_data_dir(uuid, tmp_character_model)
if not os.path.exists(folder_path):
continue
else:
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
file_lora_path_swift = f"{file_path}/swift"
if os.path.exists(file_lora_path) or os.path.exists(file_lora_path_swift):
folder_list.append(file)
if len(folder_list) == 0:
return gr.Radio.update(choices=[], value = None)
return gr.Radio.update(choices=folder_list)
def update_output_model_inpaint(uuid):
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
folder_path = join_worker_data_dir(uuid, 'ly261666/cv_portrait_model')
folder_list = ['不重绘该人物(Do not inpaint this character)']
if not os.path.exists(folder_path):
return gr.Radio.update(choices=[], value = None), gr.Dropdown.update(choices=style_list)
else:
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
file_lora_path_swift = f"{file_path}/unet"
if os.path.exists(file_lora_path) or os.path.exists(file_lora_path_swift):
folder_list.append(file)
return gr.Radio.update(choices=folder_list, value=folder_list[0]), gr.Radio.update(choices=folder_list, value=folder_list[0])
def add_file_webcam(instance_images, file):
if file is None:
instance_images = [file_d['name'] for file_d in instance_images]
return instance_images
else:
instance_images = [file_d['name'] for file_d in instance_images] + [file]
return instance_images
def webcam_image_open(image):
image = gr.update(visible=True)
return image
def webcam_image_close(image):
image = gr.update(value=None,visible=False)
return image
def update_output_model_tryon(uuid):
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
folder_path = join_worker_data_dir(uuid, 'ly261666/cv_portrait_model')
folder_list = ['不重绘该人物(Do not inpaint this character)']
if not os.path.exists(folder_path):
return gr.Radio.update(choices=[], value = None)
else:
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
if os.path.exists(file_lora_path):
folder_list.append(file)
return gr.Radio.update(choices=folder_list, value=folder_list[0])
def init_output_model_tryon(uuid):
if not uuid:
if os.getenv("MODELSCOPE_ENVIRONMENT") == 'studio':
raise gr.Error("请登陆后使用! (Please login first)")
else:
uuid = 'qw'
folder_path = join_worker_data_dir(uuid, 'ly261666/cv_portrait_model')
folder_list = ['不重绘该人物(Do not inpaint this character)']
if not os.path.exists(folder_path):
choices = []
value = None
return choices, value
else:
files = os.listdir(folder_path)
for file in files:
file_path = os.path.join(folder_path, file)
if os.path.isdir(folder_path):
file_lora_path = f"{file_path}/pytorch_lora_weights.bin"
if os.path.exists(file_lora_path):
folder_list.append(file)
return folder_list, folder_list[0]
def update_output_model_num(num_faces):
if num_faces == 1:
return gr.Radio.update(), gr.Radio.update(visible=False)
else:
return gr.Radio.update(), gr.Radio.update(visible=True)
def update_output_image_result(uuid):
image_list = get_previous_image_result(uuid)
return gr.Gallery.update(value=image_list), image_list
def upload_file(files, current_files):
file_paths = [file_d['name'] for file_d in current_files] + [file.name for file in files]
return file_paths