-
Notifications
You must be signed in to change notification settings - Fork 1
/
animeCreator.py
3712 lines (2965 loc) · 131 KB
/
animeCreator.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
from generationFunctions import GenerationFunctions
from animdiffwrapper import generateGif
from load_llama_model import getllama
import builtins
import contextlib
from text_to_phonemes import processAudio
import sys
sys.path.append('.\AAAI22-one-shot-talking-face')
from test_script import test_with_input_audio_and_image, parse_phoneme_file
from exampleScenes import exampleScenesPrompt, exampleScenesResult
from exampleChapters import examplechapterPrompt, exampleChapterResults
from example_screenplay import exampleScreenplayPrompt, exampleScreenplayResult
import datetime
import uuid
import logging
# from riffusion import get_music
# import riffusion
from worldObject import WorldObject, ListObject
from templates import templates
from mubert import generate_track_by_prompt
import IPython.display as ipd
from fairseq.models.text_to_speech.hub_interface import TTSHubInterface
from fairseq.checkpoint_utils import load_model_ensemble_and_task_from_hf_hub
import pits.app as pits
import traceback
from diffusers.models import AutoencoderKL
from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler, StableDiffusionImg2ImgPipeline, UniPCMultistepScheduler, DiffusionPipeline
from diffusers import StableDiffusionXLPipeline, AutoencoderTiny, StableDiffusionXLImg2ImgPipeline
import time
from torch import autocast
import ipywidgets as widgets
from ipywidgets import Audio # no good, doesn't stop when clear display
import numpy
import numpy as np
from io import BytesIO
from pydub import AudioSegment
import urllib
from PIL import Image, ImageFilter
import random
import torch
import gc
from transformers import AutoTokenizer, AutoModelForCausalLM
from transformers import pipeline
import re
import os
import openai
from tenacity import retry, wait_exponential, wait_combine, stop_after_attempt, after_log, before_sleep_log
from diffusers import AudioLDMPipeline
from example_classifications import example_classifications
from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
import tomesd
# from cldm.model import create_model, load_state_dict
# from cldm.ddim_hacked import DDIMSampler
from laion_face_common import generate_annotation
import subprocess
import json
import glob
from modules.sadtalker_test import SadTalker
# from multiprocessing import Pool
class CustomRootLogger(logging.Logger):
def setLevel(self, level):
stack_trace = ''.join(traceback.format_stack())
print(f"Log level changed to {level} by:\n{stack_trace}")
super().setLevel(level)
# Replace the root logger with the custom one
logging.setLoggerClass(CustomRootLogger)
root_logger = logging.getLogger()
file_handler = logging.FileHandler(filename='tmp.log')
stdout_handler = logging.StreamHandler(stream=sys.stdout)
handlers = [file_handler, stdout_handler]
logger = logging.getLogger(__name__)
logger.setLevel(logging.DEBUG) # Set the logger level
log_format = '[%(asctime)s] {%(filename)s:%(lineno)d} %(levelname)s - %(message)s'
formatter = logging.Formatter(log_format)
# Set the formatter and add handlers to the logger
for handler in handlers:
handler.setFormatter(formatter)
logger.addHandler(handler)
logger.info("logging should be working?")
def custom_exponential_wait(retry_state):
base_wait = 4
exponent = 1.2
return base_wait * (exponent ** retry_state.attempt_number)
def custom_wait_gen():
attempt = 0
while True:
yield custom_exponential_wait(attempt)
attempt += 1
# from IPython.display import Audio, display
def getFilename(path, extension):
current_datetime = datetime.datetime.now().strftime("%Y%m%d-%H%M%S")
filename = f"{path}{current_datetime}-{uuid.uuid4()}.{extension}"
return filename
class AnimeBuilder:
def __init__(
self,
textModel='EleutherAI/gpt-neo-2.7B',
diffusionModel="hakurei/waifu-diffusion",
vaeModel="stabilityai/sd-vae-ft-mse",
templates=templates,
advanceSceneObjects=None,
num_inference_steps=30,
cfg=None,
verbose=False,
doImg2Img=False,
img2imgStrength=0.4,
saveMemory=True,
cache_dir='../hf',
textRevision=None,
negativePrompt="collage, grayscale, text, watermark, lowres, bad anatomy, bad hands, text, error, missing fingers, cropped, worst quality, low quality, normal quality, jpeg artifacts, watermark, blurry, grayscale, deformed weapons, deformed face, deformed human body",
suffix=", anime drawing",
riffusionSuffix=" pleasing rythmic background music",
savePath="./static/samples/",
saveImages=False,
audioLDM="cvssp/audioldm-s-full-v2",
soundEffectDuration=1.5,
musicDuration=16,
musicSuffix=" movie soundtrack background music, smooth jazz",
imageSizes=[512, 512, 1024, 1024],
usePITS=True,
fixAsides=False,
portraitPrompt=', anime, face, portrait, headshot, white background',
computeDepth=True,
osth=True,
tokenizer=None,
use_gpt_for_chat_completion=False,
parallel_screenplays=True,
controlnet_diffusion_model="runwayml/stable-diffusion-v1-5",
video_mode=False,
blur_radius=0.5,
talking_head_decimate=1,
face_steps=20,
max_previous_scenes=6,
use_GPT4=False,
):
self.use_GPT4 = use_GPT4
self.blur_radius = blur_radius
self.max_previous_scenes = max_previous_scenes
self.talking_head_decimate = talking_head_decimate
self.face_steps = face_steps
self.saveMemory = saveMemory
self.doImg2Img = doImg2Img
# read system prompt files
self.scenePrompt = open("chapters_to_scenes_systemPrompt.txt").read()
self.chapterPrompt = open(
"summary_to_chapters_systemPrompt.txt").read()
self.screenplayPrompt = open("screenplay_systemPrompt.txt").read()
self.bonusSceneInstruction = '> NEVER UNDER ANY CIRCUMSTANCES USE THE WORD "MUST"\n\n'
# load generation functions (for now this is just img2img, move more there later)
if self.doImg2Img:
self.generationFunctions = GenerationFunctions(
saveMemory=self.saveMemory)
self.video_mode = video_mode
self.osth = osth
self.portraitPrompt = portraitPrompt
self.parallel_screenplays = parallel_screenplays
# always use parallen when using chatgpt
if use_gpt_for_chat_completion:
self.parallel_screenplays = True
self.fixAsides = fixAsides
self.imageSizes = imageSizes
self.img2imgStrength = img2imgStrength
self.soundEffectDuration = soundEffectDuration
self.musicDuration = musicDuration
self.musicSuffix = musicSuffix
self.savePath = savePath
self.saveImages = saveImages
self.use_gpt_for_chat_completion = use_gpt_for_chat_completion
self.ignored_words = set(
["the", "name", "setting", "music", "action", "sound", "effect"])
self.textModel = textModel
self.cache_dir = cache_dir
self.verbose = verbose
self.mubert = False
self.templates = templates
if cfg is None:
cfg = {
"genTextAmount_min": 30,
"genTextAmount_max": 100,
"no_repeat_ngram_size": 16,
"repetition_penalty": 1.0,
"MIN_ABC": 4,
"num_beams": 1,
"temperature": 1.0,
"MAX_DEPTH": 5
}
self.cfg = cfg
self.num_inference_steps = num_inference_steps
self.negativePrompt = negativePrompt
self.suffix = suffix
self.riffusionSuffix = riffusionSuffix
# use this for advanceScene()
# advance scene
if advanceSceneObjects is None:
self.advanceSceneObjects = [
{
"object": "advancePlot",
"whichScene": 3,
"numScenes": 3,
},
{
"object": "fightScene",
"whichScene": 1,
"numScenes": 3,
},
]
else:
self.advanceSceneObjects = advanceSceneObjects
if self.verbose:
print("LOADING TEXT MODEL")
if audioLDM is not None:
self.audioLDMPipe = AudioLDMPipeline.from_pretrained(
audioLDM, torch_dtype=torch.float16)
self.audioLDMPipe = self.audioLDMPipe.to("cuda")
# move to cpu if saving memory
if self.saveMemory:
self.audioLDMPipe = self.audioLDMPipe.to("cpu")
if self.textModel == "GPT3":
pass
# self.textGenerator="GPT3"
self.textGenerator = {
'name': "GPT3",
}
openai.organization = "org-bKm1yrKncCnPfkcf8pDpe4GM"
openai.api_key = os.getenv("OPENAI_API_KEY")
openai.Model.list()
elif self.textModel == "gpt-3.5-turbo-instruct":
# self.textGenerator="gpt-3.5-turbo-instruct"
self.textGenerator = {
'name': "gpt-3.5-turbo-instruct",
}
openai.organization = "org-bKm1yrKncCnPfkcf8pDpe4GM"
openai.api_key = os.getenv("OPENAI_API_KEY")
openai.Model.list()
elif self.textModel == "llama":
thisTokenizer, thisPipeline = getllama()
self.textGenerator = {
'name': "llama",
'tokenizer': thisTokenizer,
'pipeline': thisPipeline
}
else:
# text model
self.textModel = textModel
self.textRevision = textRevision
textGenerator = pipeline('text-generation',
torch_dtype=torch.float16,
model=self.textModel,
trust_remote_code=True,
device_map="auto",
model_kwargs={"load_in_4bit": True}
)
if tokenizer is None:
self.tokenizer = AutoTokenizer.from_pretrained(
self.textModel, torch_dtype=torch.float16)
else:
self.tokenizer = tokenizer
textGenerator.tokenizer = tokenizer
self.textGenerator = {
'name': self.textModel,
'tokenizer': self.tokenizer,
# 'model': self.textModel
'pipeline': textGenerator
}
# image model
if self.verbose:
print("LOADING IMAGE MODEL")
# make sure you're logged in with `huggingface-cli login`
# vae = AutoencoderKL.from_pretrained(vaeModel) #maybe I should enable this again?
# pipe = StableDiffusionPipeline.from_pretrained(diffusionModel,vae=vae, torch_dtype=torch.float16,custom_pipeline="composable_stable_diffusion")
# pipe = DiffusionPipeline.from_pretrained(
# diffusionModel,
# vae=vae,
# torch_dtype=torch.float16,
# custom_pipeline="lpw_stable_diffusion",
# )
self.diffusionModel = diffusionModel
if "xl" in diffusionModel.lower():
pipe = StableDiffusionXLPipeline.from_single_file(
diffusionModel, torch_dtype=torch.float16, use_safetensors=True,
custom_pipeline="lpw_stable_diffusion_xl"
)
pipe.vae = AutoencoderTiny.from_pretrained(
"madebyollin/taesdxl", torch_dtype=torch.float16)
elif diffusionModel == "LCM":
pipe = DiffusionPipeline.from_pretrained(
"SimianLuo/LCM_Dreamshaper_v7", custom_pipeline="latent_consistency_txt2img", custom_revision="main")
# To save GPU memory, torch.float16 can be used, but it may compromise image quality.
pipe.to(torch_device="cuda", torch_dtype=torch.float32)
else:
# pipe = DiffusionPipeline.from_pretrained(diffusionModel)
# check if model_id is a .ckpt or .safetensors file
if diffusionModel.endswith(".ckpt") or diffusionModel.endswith(".safetensors"):
print("about to die", diffusionModel)
pipe = StableDiffusionPipeline.from_single_file(diffusionModel,
torch_dtype=torch.float16)
else:
pipe = StableDiffusionPipeline.from_pretrained(
diffusionModel, torch_dtype=torch.float16)
# change to UniPC scheduler
pipe.scheduler = UniPCMultistepScheduler.from_config(
pipe.scheduler.config)
pipe = pipe.to("cuda")
pipe.enable_attention_slicing()
pipe.enable_xformers_memory_efficient_attention()
tomesd.apply_patch(pipe, ratio=0.5)
self.pipe = pipe
# if save memory, move pipe to cpu and do garbage collection
if self.saveMemory:
self.pipe = self.pipe.to("cpu")
gc.collect()
# collect cuda memory
torch.cuda.empty_cache()
else:
self.pipe = self.pipe.to("cuda")
self.pipe.safety_checker = None
'''
if self.doImg2Img:
if self.verbose:
print("LOADING Img2Img")
if "xl" in diffusionModel.lower():
img2img = StableDiffusionXLImg2ImgPipeline.from_single_file(
diffusionModel, torch_dtype=torch.float16, use_safetensors=True)
# img2img.vae = AutoencoderTiny.from_pretrained("madebyollin/taesdxl", torch_dtype=torch.float16)
img2img.enable_vae_tiling()
self.img2img = img2img
self.img2img.safety_checker = None
else:
if diffusionModel.endswith(".ckpt") or diffusionModel.endswith(".safetensors"):
thisModelName = "runwayml/stable-diffusion-v1-5"
else:
thisModelName = diffusionModel
self.img2img = StableDiffusionImg2ImgPipeline.from_pretrained(
thisModelName,
# revision=revision,
scheduler=self.pipe.scheduler,
unet=self.pipe.unet,
vae=self.pipe.vae,
safety_checker=self.pipe.safety_checker,
text_encoder=self.pipe.text_encoder,
tokenizer=self.pipe.tokenizer,
torch_dtype=torch.float16,
use_auth_token=True,
cache_dir="./AI/StableDiffusion"
)
self.img2img.enable_attention_slicing()
self.img2img.enable_xformers_memory_efficient_attention()
tomesd.apply_patch(self.img2img, ratio=0.5)
# if save memmory, move to cpu and do garbage collection
if self.saveMemory:
self.img2img = self.img2img.to("cpu")
gc.collect()
# collect cuda memory
torch.cuda.empty_cache()
'''
if self.verbose:
print("LOADING TTS MODEL")
# tts
#
self.usePITS = usePITS
if usePITS:
self.pitsTTS = pits.GradioApp(pits.get_default_args())
else:
models, cfg, task = load_model_ensemble_and_task_from_hf_hub(
"facebook/fastspeech2-en-200_speaker-cv4", # random
arg_overrides={"vocoder": "hifigan", "fp16": False, }
)
self.tts_models = models
self.tts_cfg = cfg
self.tts_task = task
# model = models[0]
TTSHubInterface.update_cfg_with_data_cfg(cfg, task.data_cfg)
self.tts_generator = task.build_generator(models, cfg)
# 000000000011111111112222222222333333333344444444444555555555
# 012345678901234567890123456789012345678901234567890123456789
if self.usePITS:
# 01234567890123456789012345678901234567890123456789012345678901234567890123456789012345678901234501234567890
# 00000000001111111111222222222233333333334444444444555555555566666666667777777777888888888899999999990000000000
s = "fmmffffmfffmfffmmfmmmffffmfmfmfmmmffmffffffmmmmmmffmmmffmmmmfmffffmfffmfmfffffmfffmfffmfffmffffffmmfmffmmmmf".upper()
else:
s = "FMFMMMMMFMMMFFMFFMMMMMMmffmmfmmfmfmmmmmmfmmmmmfmmmffmmmm".upper()
self.maleVoices = [i for i in range(len(s)) if s[i] == "M"]
self.femaleVoices = [i for i in range(len(s)) if s[i] == "F"]
# controlnet for portrait generation
# self.facemodel = create_model('../cldm_v21.yaml').cpu()
# self.facemodel.load_state_dict(load_state_dict(
# '..\ControlNet\models/controlnet_sd21_laion_face_v2_full.ckpt', location='cuda'))
# self.facemodel = self.facemodel.cuda()
# self.facemodel = self.facemodel.cpu()
# self.facemodel_ddim_sampler = DDIMSampler(self.facemodel) # ControlNet _only_ works with DDIM.
# Stable Diffusion 2.1-base:
# controlnet = ControlNetModel.from_pretrained(
# "CrucibleAI/ControlNetMediaPipeFace", torch_dtype=torch.float16, variant="fp16")
# self.facepipe = StableDiffusionControlNetPipeline.from_pretrained(
# "stabilityai/stable-diffusion-2-1-base", controlnet=controlnet, safety_checker=None, torch_dtype=torch.float16
# )
# controlnet = ControlNetModel.from_pretrained("CrucibleAI/ControlNetMediaPipeFace", subfolder="diffusion_sd15")
# if diffusionModel.endswith(".ckpt") or diffusionModel.endswith(".safetensors"):
# self.facepipe = StableDiffusionControlNetPipeline.from_single_file(diffusionModel, controlnet=controlnet, safety_checker=None)
# else:
# self.facepipe = StableDiffusionControlNetPipeline.from_pretrained(diffusionModel, controlnet=controlnet, safety_checker=None)
# self.facepipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
# Remove if you do not have xformers installed
# see https://huggingface.co/docs/diffusers/v0.13.0/en/optimization/xformers#installing-xformers
# for installation instructions
# self.facepipe.enable_xformers_memory_efficient_attention()
# self.facepipe.enable_model_cpu_offload()
if "xl" in diffusionModel.lower():
# TODO: add sdxl controlnet when it's available
pass
# OR
# Stable Diffusion 1.5:
controlnet = ControlNetModel.from_pretrained(
"CrucibleAI/ControlNetMediaPipeFace", subfolder="diffusion_sd15", torch_dtype=torch.float16, variant="fp16")
if "safetensors" in controlnet_diffusion_model:
self.facepipe = StableDiffusionControlNetPipeline.from_single_file(
controlnet_diffusion_model, controlnet=controlnet, safety_checker=None, torch_dtype=torch.float16)
else:
self.facepipe = StableDiffusionControlNetPipeline.from_pretrained(
controlnet_diffusion_model, controlnet=controlnet, safety_checker=None, torch_dtype=torch.float16)
# disable safety checker
self.facepipe.safety_checker = None
self.facepipe.scheduler = UniPCMultistepScheduler.from_config(
self.facepipe.scheduler.config)
# Remove if you do not have xformers installed
# see https://huggingface.co/docs/diffusers/v0.13.0/en/optimization/xformers#installing-xformers
# for installation instructions
self.facepipe.enable_xformers_memory_efficient_attention()
# self.facepipe.enable_model_cpu_offload()
# if save memmory, move to cpu and do garbage collection
if self.saveMemory:
self.facepipe = self.facepipe.to("cpu")
gc.collect()
# collect cuda memory
torch.cuda.empty_cache()
else:
self.facepipe = self.facepipe.to("cuda")
if not self.osth:
self.sad_talker = SadTalker("E:\img\SadTalker")
if computeDepth:
repo = "isl-org/ZoeDepth"
# Zoe_N
model_zoe_n = torch.hub.load(repo, "ZoeD_NK", pretrained=True)
DEVICE = "cuda"
self.zoe = model_zoe_n.to(DEVICE)
else:
self.zoe = None
if self.saveMemory:
self.zoe = self.zoe.to("cpu")
gc.collect()
# collect cuda memory
torch.cuda.empty_cache()
else:
self.zoe = self.zoe.to("cuda")
def chatCompletion(self, messages, n=1, min_new_tokens=256, max_new_tokens=512, generation_prefix=""):
# free up some memory
gc.collect()
torch.cuda.empty_cache()
# first we need to combine messages into a single string
# as a reminder messages have the format {"role": "system/user/assistant", "content": "this is some conent"}
prompt = ""
lastRole = "system"
for message in messages:
# prompt += message['role']+":\n"
if message['role'] != lastRole:
prompt += "\n"
prompt += message['content']+"\n"
lastRole = message['role']
# now add a final "assitant:" to the prompt
# prompt += "assistant:\n"
# now we can run the completion
prompt += "\n"+generation_prefix
output = []
for i in range(n):
# print("\n=====\n", prompt, "\n=====\n")
result = self.textGenerator['pipeline'](prompt,
min_new_tokens=min_new_tokens,
max_new_tokens=max_new_tokens,
return_full_text=True,
no_repeat_ngram_size=self.cfg["no_repeat_ngram_size"],
repetition_penalty=self.cfg["repetition_penalty"],
num_beams=self.cfg["num_beams"],
temperature=self.cfg["temperature"],
do_sample=True,
)
result_text = result[0]['generated_text']
# print("\n=====\n", result_text, "\n=====\n")
# now we need to pull out the resulting message
start_index = len(prompt)
# stop at \n\n
end_index = result_text.find("\n\n", start_index)
# end_index = result_text.find("user:", start_index)
# print("start_index:", start_index, "end_index:", end_index)
output += [generation_prefix+result_text[start_index:end_index]]
# output += [generation_prefix+result_text]
# free up some memory
gc.collect()
torch.cuda.empty_cache()
return output
def _get_portrait0(self, input_image: Image.Image, prompt, a_prompt, n_prompt, max_faces, num_samples, ddim_steps, guess_mode, strength, scale, seed, eta):
# move to cuda
self.facemodel = self.facemodel.cuda()
# ControlNet _only_ works with DDIM.
facemodel_ddim_sampler = DDIMSampler(self.facemodel)
with torch.no_grad():
empty = generate_annotation(input_image, max_faces)
visualization = Image.fromarray(empty) # Save to help debug.
empty = numpy.moveaxis(empty, 2, 0) # h, w, c -> c, h, w
control = torch.from_numpy(empty.copy()).float().cuda() / 255.0
control = torch.stack([control for _ in range(num_samples)], dim=0)
# control = einops.rearrange(control, 'b h w c -> b c h w').clone()
# Sanity check the dimensions.
B, C, H, W = control.shape
assert C == 3
assert B == num_samples
if seed != -1:
random.seed(seed)
os.environ['PYTHONHASHSEED'] = str(seed)
numpy.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.deterministic = True
# if config.save_memory:
# model.low_vram_shift(is_diffusing=False)
cond = {"c_concat": [control], "c_crossattn": [
self.facemodel.get_learned_conditioning([prompt + ', ' + a_prompt] * num_samples)]}
un_cond = {"c_concat": None if guess_mode else [control], "c_crossattn": [
self.facemodel.get_learned_conditioning([n_prompt] * num_samples)]}
shape = (4, H // 8, W // 8)
# if config.save_memory:
# model.low_vram_shift(is_diffusing=True)
self.facemodel.control_scales = [strength * (0.825 ** float(12 - i)) for i in range(13)] if guess_mode else (
[strength] * 13) # Magic number. IDK why. Perhaps because 0.825**12<0.01 but 0.826**12>0.01
samples, intermediates = facemodel_ddim_sampler.sample(
ddim_steps,
num_samples,
shape,
cond,
verbose=False,
eta=eta,
unconditional_guidance_scale=scale,
unconditional_conditioning=un_cond
)
# if config.save_memory:
# model.low_vram_shift(is_diffusing=False)
x_samples = self.facemodel.decode_first_stage(samples)
# x_samples = (einops.rearrange(x_samples, 'b c h w -> b h w c') * 127.5 + 127.5).cpu().numpy().clip(0, 255).astype(numpy.uint8)
x_samples = numpy.moveaxis((x_samples * 127.5 + 127.5).cpu().numpy().clip(
0, 255).astype(numpy.uint8), 1, -1) # b, c, h, w -> b, h, w, c
results = [visualization] + [x_samples[i]
for i in range(num_samples)]
# move to cpu
# move to cuda
self.facemodel = self.facemodel.to('cuda')
gc.collect()
torch.cuda.empty_cache()
return results
def _get_portrait(self, input_image: Image.Image, prompt, a_prompt, n_prompt, NUM_RETRIES=3):
empty = generate_annotation(input_image, 1)
anno = Image.fromarray(empty).resize((768, 768))
# if save memory, move from cpu to gpu
if self.saveMemory:
self.facepipe = self.facepipe.to('cuda')
image = self.facepipe(prompt+a_prompt, negative_prompt=n_prompt,
image=anno, num_inference_steps=self.face_steps).images[0]
# image = self.facepipe(prompt+a_prompt, negative_prompt=n_prompt,
# image=input_image, num_inference_steps=30).images[0]
# check if image is all black, and if so, retry
for i in range(NUM_RETRIES):
if np.all(np.array(image) == 0):
print("RETRYING PORTRAIT")
image = self.facepipe(prompt+a_prompt, negative_prompt=n_prompt,
image=anno, num_inference_steps=self.face_steps).images[0]
else:
break
# if save memory, move from gpu to cpu
if self.saveMemory:
self.facepipe = self.facepipe.to('cpu')
gc.collect()
torch.cuda.empty_cache()
image.save("./static/samples/tmp.png")
return image
def getPortrait(self, prompt, promptSuffix, img2imgStrength=0.6, num_inference_steps=20):
depth_image_path = "./nan3.jpg"
input_image = Image.open(depth_image_path)
input_image = input_image.resize((512, 512))
# a_prompt=',anime, face, portrait, headshot, white background'#added to prompt
a_prompt = self.portraitPrompt+promptSuffix
# n_prompt='longbody, lowres, bad anatomy, bad hands, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality'#negative prompt
n_prompt = "hands, watermark, "+self.negativePrompt
max_faces = 1
num_samples = 1
ddim_steps = 10
guess_mode = False
strength = 0.8
scale = 7.5 # cfg scale
seed = np.random.randint(0, 10000)
eta = 0
print("creating portrait with prompt:", prompt+a_prompt)
# results = self._get_portrait(input_image, prompt, a_prompt, n_prompt, max_faces,
# num_samples, ddim_steps, guess_mode, strength, scale, seed, eta)
# results = self._get_portrait(input_image, prompt, a_prompt, n_prompt)
# output = Image.fromarray(results[1])
output = self._get_portrait(input_image, prompt, a_prompt, n_prompt)
if self.doImg2Img:
# img2Input = output.resize((self.imageSizes[2], self.imageSizes[3]))
img2Input = output.resize((1024, 1024))
'''
# some nonsense to handle long prompts, based off of https://github.com/huggingface/diffusers/issues/2136#issuecomment-1409978949
# todo figure out what this
max_length = self.pipe.tokenizer.model_max_length
input_ids = self.pipe.tokenizer(
prompt, return_tensors="pt").input_ids
input_ids = input_ids.to("cuda")
# negative_ids = self.pipe.tokenizer(self.negativePrompt, truncation=False, padding="max_length", max_length=input_ids.shape[-1], return_tensors="pt").input_ids
negative_ids = self.pipe.tokenizer(
self.negativePrompt, truncation=True, padding="max_length", max_length=input_ids.shape[-1], return_tensors="pt").input_ids
negative_ids = negative_ids.to("cuda")
padding_length = max_length - (input_ids.shape[-1] % max_length)
if padding_length > 0:
input_ids = torch.cat([input_ids, torch.full((input_ids.shape[0], padding_length),
self.pipe.tokenizer.pad_token_id, dtype=torch.long, device="cuda")], dim=1)
negative_ids = torch.cat([negative_ids, torch.full(
(negative_ids.shape[0], padding_length), self.pipe.tokenizer.pad_token_id, dtype=torch.long, device="cuda")], dim=1)
concat_embeds = []
neg_embeds = []
for i in range(0, input_ids.shape[-1], max_length):
concat_embeds.append(self.pipe.text_encoder(
input_ids[:, i: i + max_length])[0])
neg_embeds.append(self.pipe.text_encoder(
negative_ids[:, i: i + max_length])[0])
prompt_embeds = torch.cat(concat_embeds, dim=1)
negative_prompt_embeds = torch.cat(neg_embeds, dim=1)
if self.saveMemory:
self.img2img = self.img2img.to('cuda')
# with autocast("cuda"):
if True: # for some reason autocast is bad?
img2 = self.img2img(
prompt=prompt,
negative_prompt=self.negativePrompt,
# prompt_embeds=prompt_embeds,
# negative_prompt_embeds=negative_prompt_embeds,
image=img2Input,
strength=img2imgStrength,
guidance_scale=7.5,
num_inference_steps=num_inference_steps,
).images[0]
output = img2
if self.saveMemory:
self.img2img = self.img2img.to('cpu')
gc.collect()
torch.cuda.empty_cache()
'''
img2 = self.generationFunctions.image_to_image(img2Input,
prompt,
"low resolution, blurry, "+self.negativePrompt,
img2imgStrength,
steps=num_inference_steps)
output = img2
# return output
filename = getFilename(self.savePath, "png")
output.save(filename)
if self.zoe is not None:
depthFilename = filename.replace(".png", "_depth.png")
depth = self.getZoeDepth(output)
depth.save(depthFilename)
print("DIED")
return filename
def getTalkingHeadVideo(self, portrait_image_path, text, voice, gender, supress=True, decimate=1):
audio_file_path, duration = self.textToSpeech(text, voice, gender)
# make sure audio_file_path ends with .wav (this file exists either way)
if not audio_file_path.endswith('.wav'):
audio_file_path = audio_file_path[:-4]+".wav"
if self.osth:
image_path = portrait_image_path
save_dir = getFilename(self.savePath, "mov")
if image_path.endswith('.png'):
png_path = os.path.join(image_path)
jpg_path = os.path.join(
os.path.splitext(image_path)[0] + '.jpg')
img = Image.open(png_path)
rgb_img = img.convert('RGB')
rgb_img.save(jpg_path)
image_path = jpg_path
osth_path = '.\AAAI22-one-shot-talking-face'
os.makedirs(save_dir, exist_ok=True)
phoneme = processAudio(
audio_file_path, phindex_location=".\AAAI22-one-shot-talking-face\phindex.json")
# supress printing
if supress == True:
with open(os.devnull, "w") as f, contextlib.redirect_stdout(f):
mov = test_with_input_audio_and_image(image_path, audio_file_path, phoneme,
".\\AAAI22-one-shot-talking-face\\checkpoint\\generator.ckpt",
".\\AAAI22-one-shot-talking-face\\checkpoint\\audio2pose.ckpt",
save_dir, osth_path, decimate=decimate)
else:
mov = test_with_input_audio_and_image(image_path, audio_file_path, phoneme,
".\\AAAI22-one-shot-talking-face\\checkpoint\\generator.ckpt",
".\\AAAI22-one-shot-talking-face\\checkpoint\\audio2pose.ckpt",
save_dir, osth_path, decimate=decimate)
print(mov)
found_movie = glob.glob(os.path.join(save_dir, "*.mp4"))
return found_movie[0], duration
else:
# use sadtalker
driven_audio = audio_file_path
source_image = portrait_image_path
still_mode = False
resize_mode = True
use_enhancer = False
result_dir = ".\static\samples"
result = self.sad_talker.test(
source_image,
driven_audio,
still_mode,
resize_mode,
use_enhancer,
result_dir
)
# replace all #'s with _'s in filename
newFilename = result[0].replace("#", "_")
os.rename(result[0], newFilename)
return newFilename, duration
def doGen(self, prompt, num_inference_steps=30, recursion=0):
# move text model to cpu for now
# if self.saveMemory:
# self.textGenerator['pipeline'].model = self.textGenerator['pipeline'].model.cpu(
# )
# gc.collect()
# torch.cuda.empty_cache()
seed = np.random.randint(0, 1000000)
print("SEED: ", seed, "")
generator = torch.Generator("cuda").manual_seed(seed)
print("ABOUT TO DIE")
# if save memory, move out of cpu
if self.saveMemory:
self.pipe = self.pipe.to('cuda')
if self.diffusionModel == "LCM":
image = self.pipe([prompt],
# negative_prompt=[self.negativePrompt], #not supported for some reason :(
guidance_scale=7.5,
num_inference_steps=num_inference_steps,
width=self.imageSizes[0],
height=self.imageSizes[1],
# generator=generator
).images[0]
else:
with autocast("cuda"):
image = self.pipe([prompt],
negative_prompt=[self.negativePrompt],
guidance_scale=7.5,
num_inference_steps=num_inference_steps,
width=self.imageSizes[0],
height=self.imageSizes[1],
generator=generator
).images[0]
# if save memory, move back to cpu
if self.saveMemory:
self.pipe = self.pipe.to('cpu')
gc.collect()
torch.cuda.empty_cache()
print("DIED")
image.save("./static/samples/test.png")
if self.doImg2Img:
# low pass filter
blurred_image = image.filter(
ImageFilter.GaussianBlur(radius=self.blur_radius))
img2Input = blurred_image.resize(
(self.imageSizes[2], self.imageSizes[3]))
# img2Input = image.resize((self.imageSizes[2], self.imageSizes[3]))
'''
# move img2img model to gpu for now
if self.saveMemory:
self.img2img = self.img2img.to('cuda')
# with autocast("cuda"):
if True:
img2 = self.img2img(
prompt=prompt,
negative_prompt=self.negativePrompt,