File size: 76,096 Bytes
26853cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
# Copyright 2023 Alibaba DAMO-VILAB and The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os

import inspect
from dataclasses import dataclass
from typing import Any, Dict, List, Optional, Tuple, Union

import numpy as np
import PIL
import torch
from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection

from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
from diffusers.loaders import LoraLoaderMixin
from diffusers.models import AutoencoderKL
from diffusers.models.lora import adjust_lora_scale_text_encoder
from diffusers.models.unets.unet_i2vgen_xl import I2VGenXLUNet
from diffusers.schedulers import DDIMScheduler
from diffusers.utils import (
    USE_PEFT_BACKEND,
    BaseOutput,
    logging,
    replace_example_docstring,
    scale_lora_layers,
    unscale_lora_layers,
)
from diffusers.utils.torch_utils import randn_tensor
from diffusers import DiffusionPipeline

# [Modified]
# Project import
from pnp_utils import register_time
from utils import load_ddim_latents_at_t

logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

EXAMPLE_DOC_STRING = """
    Examples:
        ```py
        >>> import torch
        >>> from diffusers import I2VGenXLPipeline

        >>> pipeline = I2VGenXLPipeline.from_pretrained("ali-vilab/i2vgen-xl", torch_dtype=torch.float16, variant="fp16")
        >>> pipeline.enable_model_cpu_offload()

        >>> image_url = "https://github.com/ali-vilab/i2vgen-xl/blob/main/data/test_images/img_0009.png?raw=true"
        >>> image = load_image(image_url).convert("RGB")

        >>> prompt = "Papers were floating in the air on a table in the library"
        >>> negative_prompt = "Distorted, discontinuous, Ugly, blurry, low resolution, motionless, static, disfigured, disconnected limbs, Ugly faces, incomplete arms"
        >>> generator = torch.manual_seed(8888)

        >>> frames = pipeline(
        ...     prompt=prompt,
        ...     image=image,
        ...     num_inference_steps=50,
        ...     negative_prompt=negative_prompt,
        ...     guidance_scale=9.0,
        ...     generator=generator
        ... ).frames[0]
        >>> video_path = export_to_gif(frames, "i2v.gif")
        ```
"""


# Copied from diffusers.pipelines.animatediff.pipeline_animatediff.tensor2vid
def tensor2vid(video: torch.Tensor, processor: "VaeImageProcessor", output_type: str = "np"):
    batch_size, channels, num_frames, height, width = video.shape
    outputs = []
    for batch_idx in range(batch_size):
        batch_vid = video[batch_idx].permute(1, 0, 2, 3)
        batch_output = processor.postprocess(batch_vid, output_type)

        outputs.append(batch_output)

    if output_type == "np":
        outputs = np.stack(outputs)

    elif output_type == "pt":
        outputs = torch.stack(outputs)

    elif not output_type == "pil":
        raise ValueError(f"{output_type} does not exist. Please choose one of ['np', 'pt', 'pil]")

    return outputs


@dataclass
class I2VGenXLPipelineOutput(BaseOutput):
    r"""
    Output class for image-to-video pipeline.

    Args:
        frames (`List[np.ndarray]` or `torch.FloatTensor`)
            List of denoised frames (essentially images) as NumPy arrays of shape `(height, width, num_channels)` or as
            a `torch` tensor. The length of the list denotes the video length (the number of frames).
    """

    frames: Union[List[np.ndarray], torch.FloatTensor]


# Modified from DiffEditInversionPipelineOutput
@dataclass
class StableVideoDiffusionInversionPipelineOutput(BaseOutput):
    """
    Output class for Stable Diffusion pipelines.

    Args:
        latents (`torch.FloatTensor`)
            inverted latents tensor
        images (`List[PIL.Image.Image]` or `np.ndarray`)
            List of denoised PIL images of length `num_timesteps * batch_size` or numpy array of shape `(num_timesteps,
            batch_size, height, width, num_channels)`. PIL images or numpy array present the denoised images of the
            diffusion pipeline.
    """

    inverted_latents: torch.FloatTensor
    # images: Union[List[PIL.Image.Image], np.ndarray] # TODO: we can return the noisy video.


class I2VGenXLPipeline(DiffusionPipeline):
    r"""
    Pipeline for image-to-video generation as proposed in [I2VGenXL](https://i2vgen-xl.github.io/).

    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
    implemented for all pipelines (downloading, saving, running on a particular device, etc.).

    Args:
        vae ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
        text_encoder ([`CLIPTextModel`]):
            Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
        tokenizer (`CLIPTokenizer`):
            A [`~transformers.CLIPTokenizer`] to tokenize text.
        unet ([`I2VGenXLUNet`]):
            A [`I2VGenXLUNet`] to denoise the encoded video latents.
        scheduler ([`DDIMScheduler`]):
            A scheduler to be used in combination with `unet` to denoise the encoded image latents.
    """

    model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"

    def __init__(
            self,
            vae: AutoencoderKL,
            text_encoder: CLIPTextModel,
            tokenizer: CLIPTokenizer,
            image_encoder: CLIPVisionModelWithProjection,
            feature_extractor: CLIPImageProcessor,
            unet: I2VGenXLUNet,
            scheduler: DDIMScheduler,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            image_encoder=image_encoder,
            feature_extractor=feature_extractor,
            unet=unet,
            scheduler=scheduler,
        )
        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
        # `do_resize=False` as we do custom resizing.
        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_resize=False)

    @property
    def guidance_scale(self):
        return self._guidance_scale

    # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
    # corresponds to doing no classifier free guidance.
    @property
    def do_classifier_free_guidance(self):
        return self._guidance_scale > 1

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing
    def enable_vae_slicing(self):
        r"""
        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
        """
        self.vae.enable_slicing()

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_vae_slicing
    def disable_vae_slicing(self):
        r"""
        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_slicing()

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_tiling
    def enable_vae_tiling(self):
        r"""
        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
        processing larger images.
        """
        self.vae.enable_tiling()

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_vae_tiling
    def disable_vae_tiling(self):
        r"""
        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
        computing decoding in one step.
        """
        self.vae.disable_tiling()

    def encode_prompt(
            self,
            prompt,
            device,
            num_videos_per_prompt,
            negative_prompt=None,
            prompt_embeds: Optional[torch.FloatTensor] = None,
            negative_prompt_embeds: Optional[torch.FloatTensor] = None,
            lora_scale: Optional[float] = None,
            clip_skip: Optional[int] = None,
    ):
        r"""
        Encodes the prompt into text encoder hidden states.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            device: (`torch.device`):
                torch device
            num_videos_per_prompt (`int`):
                number of images that should be generated per prompt
            do_classifier_free_guidance (`bool`):
                whether to use classifier free guidance or not
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
                less than `1`).
            prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
            lora_scale (`float`, *optional*):
                A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.
        """
        # set lora scale so that monkey patched LoRA
        # function of text encoder can correctly access it
        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
            self._lora_scale = lora_scale

            # dynamically adjust the LoRA scale
            if not USE_PEFT_BACKEND:
                adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
            else:
                scale_lora_layers(self.text_encoder, lora_scale)

        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        if prompt_embeds is None:
            text_inputs = self.tokenizer(
                prompt,
                padding="max_length",
                max_length=self.tokenizer.model_max_length,
                truncation=True,
                return_tensors="pt",
            )
            text_input_ids = text_inputs.input_ids
            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids

            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
                    text_input_ids, untruncated_ids
            ):
                removed_text = self.tokenizer.batch_decode(
                    untruncated_ids[:, self.tokenizer.model_max_length - 1: -1]
                )
                logger.warning(
                    "The following part of your input was truncated because CLIP can only handle sequences up to"
                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
                )

            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
                attention_mask = text_inputs.attention_mask.to(device)
            else:
                attention_mask = None

            if clip_skip is None:
                prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask)
                prompt_embeds = prompt_embeds[0]
            else:
                prompt_embeds = self.text_encoder(
                    text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True
                )
                # Access the `hidden_states` first, that contains a tuple of
                # all the hidden states from the encoder layers. Then index into
                # the tuple to access the hidden states from the desired layer.
                prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
                # We also need to apply the final LayerNorm here to not mess with the
                # representations. The `last_hidden_states` that we typically use for
                # obtaining the final prompt representations passes through the LayerNorm
                # layer.
                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)

        if self.text_encoder is not None:
            prompt_embeds_dtype = self.text_encoder.dtype
        elif self.unet is not None:
            prompt_embeds_dtype = self.unet.dtype
        else:
            prompt_embeds_dtype = prompt_embeds.dtype

        prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)

        bs_embed, seq_len, _ = prompt_embeds.shape
        # duplicate text embeddings for each generation per prompt, using mps friendly method
        prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
        prompt_embeds = prompt_embeds.view(bs_embed * num_videos_per_prompt, seq_len, -1)

        # get unconditional embeddings for classifier free guidance
        if self.do_classifier_free_guidance and negative_prompt_embeds is None:
            uncond_tokens: List[str]
            if negative_prompt is None:
                uncond_tokens = [""] * batch_size
            elif prompt is not None and type(prompt) is not type(negative_prompt):
                raise TypeError(
                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                    f" {type(prompt)}."
                )
            elif isinstance(negative_prompt, str):
                uncond_tokens = [negative_prompt]
            elif batch_size != len(negative_prompt):
                raise ValueError(
                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                    " the batch size of `prompt`."
                )
            else:
                uncond_tokens = negative_prompt

            max_length = prompt_embeds.shape[1]
            uncond_input = self.tokenizer(
                uncond_tokens,
                padding="max_length",
                max_length=max_length,
                truncation=True,
                return_tensors="pt",
            )

            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
                attention_mask = uncond_input.attention_mask.to(device)
            else:
                attention_mask = None

            # Apply clip_skip to negative prompt embeds
            if clip_skip is None:
                negative_prompt_embeds = self.text_encoder(
                    uncond_input.input_ids.to(device),
                    attention_mask=attention_mask,
                )
                negative_prompt_embeds = negative_prompt_embeds[0]
            else:
                negative_prompt_embeds = self.text_encoder(
                    uncond_input.input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True
                )
                # Access the `hidden_states` first, that contains a tuple of
                # all the hidden states from the encoder layers. Then index into
                # the tuple to access the hidden states from the desired layer.
                negative_prompt_embeds = negative_prompt_embeds[-1][-(clip_skip + 1)]
                # We also need to apply the final LayerNorm here to not mess with the
                # representations. The `last_hidden_states` that we typically use for
                # obtaining the final prompt representations passes through the LayerNorm
                # layer.
                negative_prompt_embeds = self.text_encoder.text_model.final_layer_norm(negative_prompt_embeds)

        if self.do_classifier_free_guidance:
            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
            seq_len = negative_prompt_embeds.shape[1]

            negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)

            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_videos_per_prompt, 1)
            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)

        if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
            # Retrieve the original scale by scaling back the LoRA layers
            unscale_lora_layers(self.text_encoder, lora_scale)

        return prompt_embeds, negative_prompt_embeds

    def _encode_image(self, image, device, num_videos_per_prompt):
        dtype = next(self.image_encoder.parameters()).dtype

        if not isinstance(image, torch.Tensor):
            image = self.image_processor.pil_to_numpy(image)
            image = self.image_processor.numpy_to_pt(image)

            # Normalize the image with CLIP training stats.
            image = self.feature_extractor(
                images=image,
                do_normalize=True,
                do_center_crop=False,
                do_resize=False,
                do_rescale=False,
                return_tensors="pt",
            ).pixel_values

        image = image.to(device=device, dtype=dtype)
        image_embeddings = self.image_encoder(image).image_embeds
        image_embeddings = image_embeddings.unsqueeze(1)

        # duplicate image embeddings for each generation per prompt, using mps friendly method
        bs_embed, seq_len, _ = image_embeddings.shape
        image_embeddings = image_embeddings.repeat(1, num_videos_per_prompt, 1)
        image_embeddings = image_embeddings.view(bs_embed * num_videos_per_prompt, seq_len, -1)

        if self.do_classifier_free_guidance:
            negative_image_embeddings = torch.zeros_like(image_embeddings)  # TODO: Why is this zero?
            image_embeddings = torch.cat([negative_image_embeddings, image_embeddings])

        return image_embeddings

    def decode_latents(self, latents, decode_chunk_size=None):
        latents = 1 / self.vae.config.scaling_factor * latents

        batch_size, channels, num_frames, height, width = latents.shape
        latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * num_frames, channels, height, width)

        if decode_chunk_size is not None:
            frames = []
            for i in range(0, latents.shape[0], decode_chunk_size):
                frame = self.vae.decode(latents[i: i + decode_chunk_size]).sample
                frames.append(frame)
            image = torch.cat(frames, dim=0)
        else:
            image = self.vae.decode(latents).sample

        decode_shape = (batch_size, num_frames, -1) + image.shape[2:]
        video = image[None, :].reshape(decode_shape).permute(0, 2, 1, 3, 4)

        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
        video = video.float()
        return video

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
    def prepare_extra_step_kwargs(self, generator, eta):
        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
        # and should be between [0, 1]

        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
        extra_step_kwargs = {}
        if accepts_eta:
            extra_step_kwargs["eta"] = eta

        # check if the scheduler accepts generator
        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
        if accepts_generator:
            extra_step_kwargs["generator"] = generator
        return extra_step_kwargs

    def check_inputs(
            self,
            prompt,
            image,
            height,
            width,
            negative_prompt=None,
            prompt_embeds=None,
            negative_prompt_embeds=None,
    ):
        if height % 8 != 0 or width % 8 != 0:
            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")

        if prompt is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
            )
        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if negative_prompt is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )

        if prompt_embeds is not None and negative_prompt_embeds is not None:
            if prompt_embeds.shape != negative_prompt_embeds.shape:
                raise ValueError(
                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
                    f" {negative_prompt_embeds.shape}."
                )

        if (
                not isinstance(image, torch.Tensor)
                and not isinstance(image, PIL.Image.Image)
                and not isinstance(image, list)
        ):
            raise ValueError(
                "`image` has to be of type `torch.FloatTensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is"
                f" {type(image)}"
            )

    def prepare_image_latents(
            self,
            image,
            device,
            num_frames,
            num_videos_per_prompt,
    ):
        image = image.to(device=device)
        image_latents = self.vae.encode(image).latent_dist.sample()
        image_latents = image_latents * self.vae.config.scaling_factor

        # Add frames dimension to image latents
        image_latents = image_latents.unsqueeze(2)

        # Append a position mask for each subsequent frame
        # after the intial image latent frame
        frame_position_mask = []
        for frame_idx in range(num_frames - 1):
            scale = (frame_idx + 1) / (num_frames - 1)
            frame_position_mask.append(torch.ones_like(image_latents[:, :, :1]) * scale)
        if frame_position_mask:
            frame_position_mask = torch.cat(frame_position_mask, dim=2)
            image_latents = torch.cat([image_latents, frame_position_mask], dim=2)

        # duplicate image_latents for each generation per prompt, using mps friendly method
        image_latents = image_latents.repeat(num_videos_per_prompt, 1, 1, 1, 1)

        if self.do_classifier_free_guidance:
            image_latents = torch.cat([image_latents] * 2)

        return image_latents

    # Modified from SVD/_encode_vae_image
    def encode_vae_video(
            self,
            video: List[PIL.Image.Image],
            device,
            height: int = 576,
            width: int = 1024,
    ):
        # video is a list of PIL images
        # [batch*frames] while batch is always 1  TODO: generalize to batch > 1
        dtype = next(self.vae.parameters()).dtype
        n_frames = len(video)
        video_latents = []
        for i in range(0, n_frames):
            frame = video[i]
            resized_frame = _center_crop_wide(frame, (width, height))
            frame = self.image_processor.preprocess(resized_frame)
            frame = frame.to(device=device, dtype=dtype)
            image_latents = self.vae.encode(frame).latent_dist.sample()  # [1, channels, height, width]
            image_latents = image_latents * self.vae.config.scaling_factor
            logger.debug(f"image_latents.shape: {image_latents.shape}")
            image_latents = image_latents.squeeze(0)  # [channels, height, width]
            video_latents.append(image_latents)
        video_latents = torch.stack(video_latents)  # [batch*frames, channels, height, width]
        video_latents = video_latents.reshape(1, n_frames, *video_latents.shape[1:])
        video_latents = video_latents.permute(0, 2, 1, 3, 4)  # [batch, channels, frames, height, width]
        video_latents = video_latents.to(device=device, dtype=dtype)
        # [batch, channels, frames, height, width]
        return video_latents

    # Copied from diffusers.pipelines.text_to_video_synthesis.pipeline_text_to_video_synth.TextToVideoSDPipeline.prepare_latents
    def prepare_latents(
            self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None
    ):
        shape = (
            batch_size,
            num_channels_latents,
            num_frames,
            height // self.vae_scale_factor,
            width // self.vae_scale_factor,
        )
        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )

        if latents is None:
            latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
        else:
            latents = latents.to(device)

        # scale the initial noise by the standard deviation required by the scheduler
        latents = latents * self.scheduler.init_noise_sigma
        logger.debug(f"latents.shape: {latents.shape}")
        logger.debug(f"init_noise_sigma: {self.scheduler.init_noise_sigma}")
        return latents

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_freeu
    def enable_freeu(self, s1: float, s2: float, b1: float, b2: float):
        r"""Enables the FreeU mechanism as in https://arxiv.org/abs/2309.11497.

        The suffixes after the scaling factors represent the stages where they are being applied.

        Please refer to the [official repository](https://github.com/ChenyangSi/FreeU) for combinations of the values
        that are known to work well for different pipelines such as Stable Diffusion v1, v2, and Stable Diffusion XL.

        Args:
            s1 (`float`):
                Scaling factor for stage 1 to attenuate the contributions of the skip features. This is done to
                mitigate "oversmoothing effect" in the enhanced denoising process.
            s2 (`float`):
                Scaling factor for stage 2 to attenuate the contributions of the skip features. This is done to
                mitigate "oversmoothing effect" in the enhanced denoising process.
            b1 (`float`): Scaling factor for stage 1 to amplify the contributions of backbone features.
            b2 (`float`): Scaling factor for stage 2 to amplify the contributions of backbone features.
        """
        if not hasattr(self, "unet"):
            raise ValueError("The pipeline must have `unet` for using FreeU.")
        self.unet.enable_freeu(s1=s1, s2=s2, b1=b1, b2=b2)

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_freeu
    def disable_freeu(self):
        """Disables the FreeU mechanism if enabled."""
        self.unet.disable_freeu()

    @torch.no_grad()
    @replace_example_docstring(EXAMPLE_DOC_STRING)
    def __call__(
            self,
            prompt: Union[str, List[str]] = None,
            image: PipelineImageInput = None,
            height: Optional[int] = 704,
            width: Optional[int] = 1280,
            target_fps: Optional[int] = 16,
            num_frames: int = 16,
            num_inference_steps: int = 50,
            guidance_scale: float = 9.0,
            negative_prompt: Optional[Union[str, List[str]]] = None,
            eta: float = 0.0,
            num_videos_per_prompt: Optional[int] = 1,
            decode_chunk_size: Optional[int] = 1,
            generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
            latents: Optional[torch.FloatTensor] = None,
            prompt_embeds: Optional[torch.FloatTensor] = None,
            negative_prompt_embeds: Optional[torch.FloatTensor] = None,
            output_type: Optional[str] = "pil",
            return_dict: bool = True,
            cross_attention_kwargs: Optional[Dict[str, Any]] = None,
            clip_skip: Optional[int] = 1,
            ddim_init_latents_t_idx: Optional[int] = 1,  # Modified
    ):
        r"""
        The call function to the pipeline for image-to-video generation with [`I2VGenXLPipeline`].

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
            image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `torch.FloatTensor`):
                Image or images to guide image generation. If you provide a tensor, it needs to be compatible with
                [`CLIPImageProcessor`](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json).
            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The width in pixels of the generated image.
            target_fps (`int`, *optional*):
                Frames per second. The rate at which the generated images shall be exported to a video after generation. This is also used as a "micro-condition" while generation.
            num_frames (`int`, *optional*):
                The number of video frames to generate.
            num_inference_steps (`int`, *optional*):
                The number of denoising steps.
            guidance_scale (`float`, *optional*, defaults to 7.5):
                A higher guidance scale value encourages the model to generate images closely linked to the text
                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
            eta (`float`, *optional*):
                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
            num_videos_per_prompt (`int`, *optional*):
                The number of images to generate per prompt.
            decode_chunk_size (`int`, *optional*):
                The number of frames to decode at a time. The higher the chunk size, the higher the temporal consistency
                between frames, but also the higher the memory consumption. By default, the decoder will decode all frames at once
                for maximal quality. Reduce `decode_chunk_size` to reduce memory usage.
            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                generation deterministic.
            latents (`torch.FloatTensor`, *optional*):
                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor is generated by sampling using the supplied random `generator`.
            prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                provided, text embeddings are generated from the `prompt` input argument.
            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            cross_attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.

        Examples:

        Returns:
            [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] or `tuple`:
                If `return_dict` is `True`, [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] is
                returned, otherwise a `tuple` is returned where the first element is a list with the generated frames.
        """
        # 0. Default height and width to unet
        height = height or self.unet.config.sample_size * self.vae_scale_factor
        width = width or self.unet.config.sample_size * self.vae_scale_factor
        logger.info(f"height: {height}, width: {width}")

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(prompt, image, height, width, negative_prompt, prompt_embeds, negative_prompt_embeds)
        logger.info(f"Prompt: {prompt}")

        # 2. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        device = self._execution_device
        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
        # corresponds to doing no classifier free guidance.
        self._guidance_scale = guidance_scale

        # 3.1 Encode input text prompt
        text_encoder_lora_scale = (
            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
        )
        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
            prompt,
            device,
            num_videos_per_prompt,
            negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            lora_scale=text_encoder_lora_scale,
            clip_skip=clip_skip,
        )
        # For classifier free guidance, we need to do two forward passes.
        # Here we concatenate the unconditional and text embeddings into a single batch
        # to avoid doing two forward passes
        if self.do_classifier_free_guidance:
            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])

        # 3.2 Encode image prompt
        # 3.2.1 Image encodings.
        # https://github.com/ali-vilab/i2vgen-xl/blob/2539c9262ff8a2a22fa9daecbfd13f0a2dbc32d0/tools/inferences/inference_i2vgen_entrance.py#L114
        cropped_image = _center_crop_wide(image, (width, width))
        cropped_image = _resize_bilinear(
            cropped_image, (self.feature_extractor.crop_size["width"], self.feature_extractor.crop_size["height"])
        )
        image_embeddings = self._encode_image(cropped_image, device, num_videos_per_prompt)

        # 3.2.2 Image latents.
        resized_image = _center_crop_wide(image, (width, height))
        image = self.image_processor.preprocess(resized_image).to(device=device, dtype=image_embeddings.dtype)
        image_latents = self.prepare_image_latents(
            image,
            device=device,
            num_frames=num_frames,
            num_videos_per_prompt=num_videos_per_prompt,
        )

        # 3.3 Prepare additional conditions for the UNet.
        if self.do_classifier_free_guidance:
            fps_tensor = torch.tensor([target_fps, target_fps]).to(device)
        else:
            fps_tensor = torch.tensor([target_fps]).to(device)
        fps_tensor = fps_tensor.repeat(batch_size * num_videos_per_prompt, 1).ravel()

        # 4. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=device)
        self.scheduler.timesteps = self.scheduler.timesteps[ddim_init_latents_t_idx:]
        timesteps = self.scheduler.timesteps
        logger.info(f"self.scheduler: {self.scheduler}")
        logger.info(f"timesteps: {timesteps}")
        logger.info(f"Sampling starts from latents_at_t={self.scheduler.timesteps[0]}")

        # 5. Prepare latent variables
        num_channels_latents = self.unet.config.in_channels
        latents = self.prepare_latents(
            batch_size * num_videos_per_prompt,
            num_channels_latents,
            num_frames,
            height,
            width,
            prompt_embeds.dtype,
            device,
            generator,
            latents,
        )

        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

        # 7. Denoising loop
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                # expand the latents if we are doing classifier free guidance
                latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=prompt_embeds,
                    fps=fps_tensor,
                    image_latents=image_latents,
                    image_embeddings=image_embeddings,
                    cross_attention_kwargs=cross_attention_kwargs,
                    return_dict=False,
                )[0]

                # perform guidance
                if self.do_classifier_free_guidance:
                    logger.debug(f"doing classifier free guidance with guidance_scale: {guidance_scale}")
                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

                # reshape latents
                batch_size, channel, frames, width, height = latents.shape
                latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)
                noise_pred = noise_pred.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample

                # reshape latents back
                latents = latents[None, :].reshape(batch_size, frames, channel, width, height).permute(0, 2, 1, 3, 4)
                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        if output_type == "latent":
            return I2VGenXLPipelineOutput(frames=latents)

        video_tensor = self.decode_latents(latents, decode_chunk_size=decode_chunk_size)
        video = tensor2vid(video_tensor, self.image_processor, output_type=output_type)

        # Offload all models
        self.maybe_free_model_hooks()

        if not return_dict:
            return (video,)

        return I2VGenXLPipelineOutput(frames=video)

    @torch.no_grad()
    @replace_example_docstring(EXAMPLE_DOC_STRING)
    def sample_with_pnp(
            self,
            prompt: Union[str, List[str]] = None,
            image: PipelineImageInput = None,
            height: Optional[int] = 704,
            width: Optional[int] = 1280,
            target_fps: Optional[int] = 16,
            num_frames: int = 16,
            num_inference_steps: int = 50,
            guidance_scale: float = 9.0,
            negative_prompt: Optional[Union[str, List[str]]] = None,
            eta: float = 0.0,
            num_videos_per_prompt: Optional[int] = 1,
            decode_chunk_size: Optional[int] = 1,
            generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
            latents: Optional[torch.FloatTensor] = None,
            prompt_embeds: Optional[torch.FloatTensor] = None,
            negative_prompt_embeds: Optional[torch.FloatTensor] = None,
            output_type: Optional[str] = "pil",
            return_dict: bool = True,
            cross_attention_kwargs: Optional[Dict[str, Any]] = None,
            clip_skip: Optional[int] = 1,
            ddim_init_latents_t_idx: Optional[int] = 1,  # [Modified]
            ddim_inv_latents_path: Optional[str] = None,  # [Modified]
            ddim_inv_prompt: Union[str, List[str]] = None,  # [Modified] this is the same prompt for ddim reconstruction
            ddim_inv_1st_frame: PipelineImageInput = None, # [Modified]
    ):
        r"""
        The call function to the pipeline for image-to-video generation with [`I2VGenXLPipeline`].

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
            image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `torch.FloatTensor`):
                Image or images to guide image generation. If you provide a tensor, it needs to be compatible with
                [`CLIPImageProcessor`](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json).
            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The width in pixels of the generated image.
            target_fps (`int`, *optional*):
                Frames per second. The rate at which the generated images shall be exported to a video after generation. This is also used as a "micro-condition" while generation.
            num_frames (`int`, *optional*):
                The number of video frames to generate.
            num_inference_steps (`int`, *optional*):
                The number of denoising steps.
            guidance_scale (`float`, *optional*, defaults to 7.5):
                A higher guidance scale value encourages the model to generate images closely linked to the text
                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
            eta (`float`, *optional*):
                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
            num_videos_per_prompt (`int`, *optional*):
                The number of images to generate per prompt.
            decode_chunk_size (`int`, *optional*):
                The number of frames to decode at a time. The higher the chunk size, the higher the temporal consistency
                between frames, but also the higher the memory consumption. By default, the decoder will decode all frames at once
                for maximal quality. Reduce `decode_chunk_size` to reduce memory usage.
            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                generation deterministic.
            latents (`torch.FloatTensor`, *optional*):
                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor is generated by sampling using the supplied random `generator`.
            prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                provided, text embeddings are generated from the `prompt` input argument.
            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            cross_attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.

        Examples:

        Returns:
            [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] or `tuple`:
                If `return_dict` is `True`, [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] is
                returned, otherwise a `tuple` is returned where the first element is a list with the generated frames.
        """
        # 0. Default height and width to unet
        height = height or self.unet.config.sample_size * self.vae_scale_factor
        width = width or self.unet.config.sample_size * self.vae_scale_factor
        logger.info(f"height: {height}, width: {width}")

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(prompt, image, height, width, negative_prompt, prompt_embeds, negative_prompt_embeds)
        logger.info(f"Prompt: {prompt}")

        # 2. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
            # [Modified]
            assert len(ddim_inv_prompt) == len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        device = self._execution_device
        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
        # corresponds to doing no classifier free guidance.
        self._guidance_scale = guidance_scale

        # 3.1 Encode input text prompt
        text_encoder_lora_scale = (
            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
        )
        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
            prompt,
            device,
            num_videos_per_prompt,
            negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            lora_scale=text_encoder_lora_scale,
            clip_skip=clip_skip,
        )

        # [Modified]
        # 3.1 Encode ddim inversion prompt
        ddim_inv_prompt_embeds, _ = self.encode_prompt(
            ddim_inv_prompt,
            device,
            num_videos_per_prompt,
            negative_prompt=None,
            prompt_embeds=None,
            negative_prompt_embeds=None,
            lora_scale=text_encoder_lora_scale,
            clip_skip=clip_skip,
        )

        # [Modified]
        # For classifier free guidance, we need to do two forward passes.
        # Here we concatenate the unconditional and text embeddings into a single batch
        # to avoid doing two forward passes
        # [ddim_inversion_prompt, editing_negative_prompt, editing_prompt]
        if self.do_classifier_free_guidance:
            prompt_embeds_all = torch.cat([ddim_inv_prompt_embeds, negative_prompt_embeds, prompt_embeds])
        else:
            prompt_embeds_all = torch.cat([ddim_inv_prompt_embeds, prompt_embeds])

        # 3.2 Encode image prompt
        # 3.2.1 Image encodings.
        # https://github.com/ali-vilab/i2vgen-xl/blob/2539c9262ff8a2a22fa9daecbfd13f0a2dbc32d0/tools/inferences/inference_i2vgen_entrance.py#L114
        cropped_image = _center_crop_wide(image, (width, width))
        cropped_image = _resize_bilinear(
            cropped_image, (self.feature_extractor.crop_size["width"], self.feature_extractor.crop_size["height"])
        )
        image_embeddings = self._encode_image(cropped_image, device, num_videos_per_prompt)

        # 3.2.2 Image latents.
        resized_image = _center_crop_wide(image, (width, height))
        image = self.image_processor.preprocess(resized_image).to(device=device, dtype=image_embeddings.dtype)
        image_latents = self.prepare_image_latents(
            image,
            device=device,
            num_frames=num_frames,
            num_videos_per_prompt=num_videos_per_prompt,
        )

        # [Modified]
        # 3.2.1 Edited first frame encodings.
        cropped_image = _center_crop_wide(ddim_inv_1st_frame, (width, width))
        cropped_image = _resize_bilinear(
            cropped_image, (self.feature_extractor.crop_size["width"], self.feature_extractor.crop_size["height"])
        )
        _embeddings = self._encode_image(cropped_image, device, num_videos_per_prompt)
        if self.do_classifier_free_guidance:
            ddim_inv_1st_frame_embeddings = _embeddings.chunk(2)[1] # Chunk 0 is negative prompt
        else:
            ddim_inv_1st_frame_embeddings = _embeddings

        # 3.2.2 Edited first frame latents.
        resized_image = _center_crop_wide(ddim_inv_1st_frame, (width, height))
        ddim_inv_1st_frame = self.image_processor.preprocess(resized_image).to(device=device, dtype=image_embeddings.dtype)
        _latents = self.prepare_image_latents(
            ddim_inv_1st_frame,
            device=device,
            num_frames=num_frames,
            num_videos_per_prompt=num_videos_per_prompt,
        )
        if self.do_classifier_free_guidance:
            ddim_inv_1st_frame_latents = _latents.chunk(2)[1]
        else:
            ddim_inv_1st_frame_latents = _latents

        image_embeddings_all = torch.cat([ddim_inv_1st_frame_embeddings, image_embeddings])
        image_latents_all = torch.cat([ddim_inv_1st_frame_latents, image_latents])

        # 3.3 Prepare additional conditions for the UNet.
        if self.do_classifier_free_guidance:
            fps_tensor = torch.tensor([target_fps, target_fps, target_fps]).to(device)
        else:
            fps_tensor = torch.tensor([target_fps, target_fps]).to(device)
        fps_tensor = fps_tensor.repeat(batch_size * num_videos_per_prompt, 1).ravel()

        # 4. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=device)
        self.scheduler.timesteps = self.scheduler.timesteps[ddim_init_latents_t_idx:]
        timesteps = self.scheduler.timesteps
        logger.info(f"self.scheduler: {self.scheduler}")
        logger.info(f"timesteps: {timesteps}")
        logger.info(f"Sampling starts from latents_at_t={self.scheduler.timesteps[0]}")

        # 5. Prepare latent variables
        num_channels_latents = self.unet.config.in_channels
        latents = self.prepare_latents(
            batch_size * num_videos_per_prompt,
            num_channels_latents,
            num_frames,
            height,
            width,
            prompt_embeds.dtype,
            device,
            generator,
            latents,
        )

        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

        # 7. Denoising loop
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                # [Modified]
                # Order: ddim_inversion_prompt, editing_negative_prompt, editing_prompt
                ddim_inv_latents_at_t = load_ddim_latents_at_t(t, ddim_inv_latents_path).to(self.device)
                if self.do_classifier_free_guidance:
                    latent_model_input = torch.cat([ddim_inv_latents_at_t, latents, latents])
                else:
                    latent_model_input = torch.cat([ddim_inv_latents_at_t, latents])
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

                # [Modified]
                # Pnp
                register_time(self, t.item())

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=prompt_embeds_all,
                    fps=fps_tensor,
                    image_latents=image_latents_all,
                    image_embeddings=image_embeddings_all,
                    cross_attention_kwargs=cross_attention_kwargs,
                    return_dict=False,
                )[0]

                # [Modified]
                # Order: ddim_inversion_prompt, editing_negative_prompt, editing_prompt
                if self.do_classifier_free_guidance:
                    _noise_pred_ddim_inv, noise_pred_negative, noise_pred_editing  = noise_pred.chunk(3)
                    logger.debug(f"doing classifier free guidance with guidance_scale: {guidance_scale}")
                    noise_pred = noise_pred_negative + guidance_scale * (noise_pred_editing - noise_pred_negative)
                else:
                    _noise_pred_ddim_inv, noise_pred_editing = noise_pred.chunk(2)
                    noise_pred = noise_pred_editing

                # reshape latents
                batch_size, channel, frames, width, height = latents.shape
                latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)
                noise_pred = noise_pred.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample

                # reshape latents back
                latents = latents[None, :].reshape(batch_size, frames, channel, width, height).permute(0, 2, 1, 3, 4)
                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        if output_type == "latent":
            return I2VGenXLPipelineOutput(frames=latents)

        video_tensor = self.decode_latents(latents, decode_chunk_size=decode_chunk_size)
        video = tensor2vid(video_tensor, self.image_processor, output_type=output_type)

        # Offload all models
        self.maybe_free_model_hooks()

        if not return_dict:
            return (video,)

        return I2VGenXLPipelineOutput(frames=video)

    @torch.no_grad()
    @replace_example_docstring(EXAMPLE_DOC_STRING)
    def invert(
            self,
            prompt: Union[str, List[str]] = None,
            image: PipelineImageInput = None,
            height: Optional[int] = 704,
            width: Optional[int] = 1280,
            target_fps: Optional[int] = 16,
            num_frames: int = 16,
            num_inference_steps: int = 50,
            guidance_scale: float = 9.0,
            negative_prompt: Optional[Union[str, List[str]]] = None,
            eta: float = 0.0,
            num_videos_per_prompt: Optional[int] = 1,
            decode_chunk_size: Optional[int] = 1,
            generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
            latents: Optional[torch.FloatTensor] = None,
            prompt_embeds: Optional[torch.FloatTensor] = None,
            negative_prompt_embeds: Optional[torch.FloatTensor] = None,
            output_type: Optional[str] = "pil",
            return_dict: bool = True,
            cross_attention_kwargs: Optional[Dict[str, Any]] = None,
            clip_skip: Optional[int] = 1,
            output_dir: Optional[str] = None,
    ):
        r"""
        The call function to the pipeline for image-to-video generation with [`I2VGenXLPipeline`].

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
            image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `torch.FloatTensor`):
                Image or images to guide image generation. If you provide a tensor, it needs to be compatible with
                [`CLIPImageProcessor`](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json).
            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The width in pixels of the generated image.
            target_fps (`int`, *optional*):
                Frames per second. The rate at which the generated images shall be exported to a video after generation. This is also used as a "micro-condition" while generation.
            num_frames (`int`, *optional*):
                The number of video frames to generate.
            num_inference_steps (`int`, *optional*):
                The number of denoising steps.
            guidance_scale (`float`, *optional*, defaults to 7.5):
                A higher guidance scale value encourages the model to generate images closely linked to the text
                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
            eta (`float`, *optional*):
                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
            num_videos_per_prompt (`int`, *optional*):
                The number of images to generate per prompt.
            decode_chunk_size (`int`, *optional*):
                The number of frames to decode at a time. The higher the chunk size, the higher the temporal consistency
                between frames, but also the higher the memory consumption. By default, the decoder will decode all frames at once
                for maximal quality. Reduce `decode_chunk_size` to reduce memory usage.
            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                generation deterministic.
            latents (`torch.FloatTensor`, *optional*):
                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor is generated by sampling using the supplied random `generator`.
            prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                provided, text embeddings are generated from the `prompt` input argument.
            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            cross_attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.

        Examples:

        Returns:
            [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] or `tuple`:
                If `return_dict` is `True`, [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] is
                returned, otherwise a `tuple` is returned where the first element is a list with the generated frames.
        """
        # 0. Default height and width to unet
        height = height or self.unet.config.sample_size * self.vae_scale_factor
        width = width or self.unet.config.sample_size * self.vae_scale_factor
        logger.info(f"height: {height}, width: {width}")

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(prompt, image, height, width, negative_prompt, prompt_embeds, negative_prompt_embeds)

        # 2. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        logger.info(f"prompt: {prompt}")

        device = self._execution_device
        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
        # corresponds to doing no classifier free guidance.
        self._guidance_scale = guidance_scale
        logger.debug(f"self._guidance_scale: {self._guidance_scale}")

        # 3.1 Encode input text prompt
        text_encoder_lora_scale = (
            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
        )
        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
            prompt,
            device,
            num_videos_per_prompt,
            negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            lora_scale=text_encoder_lora_scale,
            clip_skip=clip_skip,
        )
        # For classifier free guidance, we need to do two forward passes.
        # Here we concatenate the unconditional and text embeddings into a single batch
        # to avoid doing two forward passes
        if self.do_classifier_free_guidance:
            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])

        # 3.2 Encode image prompt
        # 3.2.1 Image encodings.
        # https://github.com/ali-vilab/i2vgen-xl/blob/2539c9262ff8a2a22fa9daecbfd13f0a2dbc32d0/tools/inferences/inference_i2vgen_entrance.py#L114
        cropped_image = _center_crop_wide(image, (width, width))
        cropped_image = _resize_bilinear(
            cropped_image, (self.feature_extractor.crop_size["width"], self.feature_extractor.crop_size["height"])
        )
        image_embeddings = self._encode_image(cropped_image, device, num_videos_per_prompt)

        # 3.2.2 Image latents.
        resized_image = _center_crop_wide(image, (width, height))
        image = self.image_processor.preprocess(resized_image).to(device=device, dtype=image_embeddings.dtype)
        image_latents = self.prepare_image_latents(
            image,
            device=device,
            num_frames=num_frames,
            num_videos_per_prompt=num_videos_per_prompt,
        )

        # 3.3 Prepare additional conditions for the UNet.
        if self.do_classifier_free_guidance:
            fps_tensor = torch.tensor([target_fps, target_fps]).to(device)
        else:
            fps_tensor = torch.tensor([target_fps]).to(device)
        fps_tensor = fps_tensor.repeat(batch_size * num_videos_per_prompt, 1).ravel()

        # 4. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=device)
        timesteps = self.scheduler.timesteps
        logger.debug(f"self.scheduler: {self.scheduler}")
        logger.debug(f"timesteps: {timesteps}")

        # 5. Prepare latent variables
        num_channels_latents = self.unet.config.in_channels
        latents = self.prepare_latents(
            batch_size * num_videos_per_prompt,
            num_channels_latents,
            num_frames,
            height,
            width,
            prompt_embeds.dtype,
            device,
            generator,
            latents,
        )

        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

        # 7. Denoising loop
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        inverted_latents = []  # Modified
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                # expand the latents if we are doing classifier free guidance
                latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

                # Concatenate image_latents over channels dimention
                logger.debug(f"image_latents.shape: {image_latents.shape}")
                logger.debug(f"latent_model_input.shape: {latent_model_input.shape}")

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=prompt_embeds,
                    fps=fps_tensor,
                    image_latents=image_latents,
                    image_embeddings=image_embeddings,
                    cross_attention_kwargs=cross_attention_kwargs,
                    return_dict=False,
                )[0]

                # perform guidance
                if self.do_classifier_free_guidance:
                    logger.debug(f"do_classifier_free_guidance with guidance_scale: {guidance_scale}")
                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)

                # reshape latents
                batch_size, channel, frames, width, height = latents.shape
                latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)
                noise_pred = noise_pred.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
                # reshape latents back
                latents = latents[None, :].reshape(batch_size, frames, channel, width, height).permute(0, 2, 1, 3, 4)

                inverted_latents.append(latents.detach().clone())  # Modified

                os.makedirs(output_dir, exist_ok=True)
                torch.save(
                    latents.detach().clone(),
                    os.path.join(output_dir, f"ddim_latents_{t}.pt"),
                )
                logger.info(f"saved noisy latents at t={t} to {output_dir}")

                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        # assert len(inverted_latents) == len(timesteps)
        inverted_latents = torch.stack(list(reversed(inverted_latents)), 1)

        if not return_dict:
            return inverted_latents

        # if output_type == "latent":
        #     return I2VGenXLPipelineOutput(frames=latents)
        #
        # video_tensor = self.decode_latents(latents, decode_chunk_size=decode_chunk_size)
        # video = tensor2vid(video_tensor, self.image_processor, output_type=output_type)

        # Offload all models
        self.maybe_free_model_hooks()

        # TODO: we can return the noisy video.
        return StableVideoDiffusionInversionPipelineOutput(inverted_latents=inverted_latents)


# The following utilities are taken and adapted from
# https://github.com/ali-vilab/i2vgen-xl/blob/main/utils/transforms.py.


def _convert_pt_to_pil(image: Union[torch.Tensor, List[torch.Tensor]]):
    if isinstance(image, list) and isinstance(image[0], torch.Tensor):
        image = torch.cat(image, 0)

    if isinstance(image, torch.Tensor):
        if image.ndim == 3:
            image = image.unsqueeze(0)

        image_numpy = VaeImageProcessor.pt_to_numpy(image)
        image_pil = VaeImageProcessor.numpy_to_pil(image_numpy)
        image = image_pil

    return image


def _resize_bilinear(
        image: Union[torch.Tensor, List[torch.Tensor], PIL.Image.Image, List[PIL.Image.Image]],
        resolution: Tuple[int, int]
):
    # First convert the images to PIL in case they are float tensors (only relevant for tests now).
    image = _convert_pt_to_pil(image)

    if isinstance(image, list):
        image = [u.resize(resolution, PIL.Image.BILINEAR) for u in image]
    else:
        image = image.resize(resolution, PIL.Image.BILINEAR)
    return image


def _center_crop_wide(
        image: Union[torch.Tensor, List[torch.Tensor], PIL.Image.Image, List[PIL.Image.Image]],
        resolution: Tuple[int, int]
):
    # First convert the images to PIL in case they are float tensors (only relevant for tests now).
    image = _convert_pt_to_pil(image)

    if isinstance(image, list):
        scale = min(image[0].size[0] / resolution[0], image[0].size[1] / resolution[1])
        image = [u.resize((round(u.width // scale), round(u.height // scale)), resample=PIL.Image.BOX) for u in image]

        # center crop
        x1 = (image[0].width - resolution[0]) // 2
        y1 = (image[0].height - resolution[1]) // 2
        image = [u.crop((x1, y1, x1 + resolution[0], y1 + resolution[1])) for u in image]
        return image
    else:
        scale = min(image.size[0] / resolution[0], image.size[1] / resolution[1])
        image = image.resize((round(image.width // scale), round(image.height // scale)), resample=PIL.Image.BOX)
        x1 = (image.width - resolution[0]) // 2
        y1 = (image.height - resolution[1]) // 2
        image = image.crop((x1, y1, x1 + resolution[0], y1 + resolution[1]))
        return image