Skip to content

unitorch.models.diffusers¤

StableProcessor¤

Bases: HfTextClassificationProcessor

Source code in src/unitorch/models/diffusers/processing_stable.py 
 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
def __init__(
    self,
    vocab_path: str,
    merge_path: str,
    vae_config_path: Optional[str] = None,
    max_seq_length: Optional[int] = 77,
    position_start_id: Optional[int] = 0,
    pad_token: Optional[str] = "<|endoftext|>",
    image_size: Optional[Tuple[int, int]] = None,
    center_crop: Optional[bool] = False,
    random_flip: Optional[bool] = False,
):
    tokenizer = CLIPTokenizer(
        vocab_file=vocab_path,
        merges_file=merge_path,
    )

    tokenizer.cls_token = tokenizer.bos_token
    tokenizer.sep_token = tokenizer.eos_token
    tokenizer.pad_token = pad_token

    HfTextClassificationProcessor.__init__(
        self,
        tokenizer=tokenizer,
        max_seq_length=max_seq_length,
        position_start_id=position_start_id,
    )

    if image_size is not None:
        self.image_size = (
            image_size
            if isinstance(image_size, tuple)
            else (image_size, image_size)
        )
    else:
        self.image_size = None
    if self.image_size is not None:
        self.vision_processor = Compose(
            [
                Resize((self.image_size[1], self.image_size[0])),
                CenterCrop((self.image_size[1], self.image_size[0]))
                if center_crop
                else RandomCrop((self.image_size[1], self.image_size[0])),
                RandomHorizontalFlip() if random_flip else Lambda(lambda x: x),
                ToTensor(),
                Normalize([0.5], [0.5]),
            ]
        )
        self.condition_vision_processor = Compose(
            [
                Resize((self.image_size[1], self.image_size[0])),
                CenterCrop((self.image_size[1], self.image_size[0])),
                ToTensor(),
            ]
        )
    else:
        self.vision_processor = Compose(
            [
                RandomHorizontalFlip() if random_flip else Lambda(lambda x: x),
                ToTensor(),
                Normalize([0.5], [0.5]),
            ]
        )
        self.condition_vision_processor = Compose(
            [
                ToTensor(),
            ]
        )

    if vae_config_path is not None:
        vae_config_dict = json.load(open(vae_config_path))
        vae_scale_factor = 2 ** (
            len(vae_config_dict.get("block_out_channels", [])) - 1
        )
        self.vae_image_processor = VaeImageProcessor(
            vae_scale_factor=vae_scale_factor
        )
        self.vae_condition_image_processor = VaeImageProcessor(
            vae_scale_factor=vae_scale_factor,
            do_convert_rgb=True,
            do_normalize=False,
        )
    else:
        self.vae_image_processor = None
        self.vae_condition_image_processor = None

StableForText2ImageGeneration¤

Bases: GenericStableModel

Source code in src/unitorch/models/diffusers/modeling_stable.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    vae_config_path: str,
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        vae_config_path=vae_config_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )

    self.pipeline = StableDiffusionPipeline(
        vae=self.vae,
        text_encoder=self.text,
        unet=self.unet,
        scheduler=self.scheduler,
        tokenizer=None,
        safety_checker=None,
        feature_extractor=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

StableForImage2ImageGeneration¤

Bases: GenericStableModel

Source code in src/unitorch/models/diffusers/modeling_stable.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    vae_config_path: str,
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        vae_config_path=vae_config_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )

    self.pipeline = StableDiffusionImg2ImgPipeline(
        vae=self.vae,
        text_encoder=self.text,
        unet=self.unet,
        scheduler=self.scheduler,
        tokenizer=None,
        safety_checker=None,
        feature_extractor=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

StableForImageInpainting¤

Bases: GenericStableModel

Source code in src/unitorch/models/diffusers/modeling_stable.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    vae_config_path: str,
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        vae_config_path=vae_config_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )

    self.pipeline = StableDiffusionInpaintPipeline(
        vae=self.vae,
        text_encoder=self.text,
        unet=self.unet,
        scheduler=self.scheduler,
        tokenizer=None,
        safety_checker=None,
        feature_extractor=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)
    self.num_channels_unet = self.unet.config.in_channels

StableForImageResolution¤

Bases: GenericStableModel

Source code in src/unitorch/models/diffusers/modeling_stable.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    vae_config_path: str,
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        vae_config_path=vae_config_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )

    self.pipeline = StableDiffusionUpscalePipeline(
        vae=self.vae,
        text_encoder=self.text,
        unet=self.unet,
        scheduler=self.scheduler,
        low_res_scheduler=self.scheduler,
        tokenizer=None,
        safety_checker=None,
        feature_extractor=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

StableXLProcessor¤

Source code in src/unitorch/models/diffusers/processing_stable_xl.py 
 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
def __init__(
    self,
    vocab_path: str,
    merge_path: str,
    vocab2_path: str,
    merge2_path: str,
    vae_config_path: Optional[str] = None,
    max_seq_length: Optional[int] = 77,
    position_start_id: Optional[int] = 0,
    pad_token: Optional[str] = "<|endoftext|>",
    pad_token2: Optional[str] = "!",
    image_size: Optional[Tuple[int, int]] = None,
    center_crop: Optional[bool] = False,
    random_flip: Optional[bool] = False,
):
    tokenizer1 = CLIPTokenizer(
        vocab_file=vocab_path,
        merges_file=merge_path,
    )

    tokenizer1.cls_token = tokenizer1.bos_token
    tokenizer1.sep_token = tokenizer1.eos_token
    tokenizer1.pad_token = pad_token

    self.text_processor1 = HfTextClassificationProcessor(
        tokenizer=tokenizer1,
        max_seq_length=max_seq_length,
        position_start_id=position_start_id,
    )

    tokenizer2 = CLIPTokenizer(
        vocab_file=vocab2_path,
        merges_file=merge2_path,
    )

    tokenizer2.cls_token = tokenizer2.bos_token
    tokenizer2.sep_token = tokenizer2.eos_token
    tokenizer2.pad_token = pad_token2

    self.text_processor2 = HfTextClassificationProcessor(
        tokenizer=tokenizer2,
        max_seq_length=max_seq_length,
        position_start_id=position_start_id,
    )

    if image_size is not None:
        self.image_size = (
            image_size
            if isinstance(image_size, tuple)
            else (image_size, image_size)
        )
    else:
        self.image_size = None

    if self.image_size is not None:
        self.vision_processor = Compose(
            [
                Resize((self.image_size[1], self.image_size[0])),
                CenterCrop((self.image_size[1], self.image_size[0]))
                if center_crop
                else RandomCrop((self.image_size[1], self.image_size[0])),
                RandomHorizontalFlip() if random_flip else Lambda(lambda x: x),
                ToTensor(),
                Normalize([0.5], [0.5]),
            ]
        )

        self.condition_vision_processor = Compose(
            [
                Resize((self.image_size[1], self.image_size[0])),
                CenterCrop((self.image_size[1], self.image_size[0])),
                ToTensor(),
            ]
        )
    else:
        self.vision_processor = Compose(
            [
                RandomHorizontalFlip() if random_flip else Lambda(lambda x: x),
                ToTensor(),
                Normalize([0.5], [0.5]),
            ]
        )

        self.condition_vision_processor = Compose(
            [
                ToTensor(),
            ]
        )

    if vae_config_path is not None:
        vae_config_dict = json.load(open(vae_config_path))
        vae_scale_factor = 2 ** (
            len(vae_config_dict.get("block_out_channels", [])) - 1
        )
        self.vae_image_processor = VaeImageProcessor(
            vae_scale_factor=vae_scale_factor
        )
        self.vae_condition_image_processor = VaeImageProcessor(
            vae_scale_factor=vae_scale_factor,
            do_convert_rgb=True,
            do_normalize=False,
        )
    else:
        self.vae_image_processor = None
        self.vae_condition_image_processor = None

StableXLForText2ImageGeneration¤

Bases: GenericStableXLModel

Source code in src/unitorch/models/diffusers/modeling_stable_xl.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    text2_config_path: str,
    vae_config_path: str,
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        text2_config_path=text2_config_path,
        vae_config_path=vae_config_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )

    self.pipeline = StableDiffusionXLPipeline(
        vae=self.vae,
        text_encoder=self.text,
        text_encoder_2=self.text2,
        unet=self.unet,
        scheduler=self.scheduler,
        tokenizer=None,
        tokenizer_2=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

StableXLForImage2ImageGeneration¤

Bases: GenericStableXLModel

Source code in src/unitorch/models/diffusers/modeling_stable_xl.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    text2_config_path: str,
    vae_config_path: str,
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        text2_config_path=text2_config_path,
        vae_config_path=vae_config_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )

    self.pipeline = StableDiffusionXLImg2ImgPipeline(
        vae=self.vae,
        text_encoder=self.text,
        text_encoder_2=self.text2,
        unet=self.unet,
        scheduler=self.scheduler,
        tokenizer=None,
        tokenizer_2=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

StableXLForImageInpainting¤

Bases: GenericStableXLModel

Source code in src/unitorch/models/diffusers/modeling_stable_xl.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    text2_config_path: str,
    vae_config_path: str,
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        text2_config_path=text2_config_path,
        vae_config_path=vae_config_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )

    self.pipeline = StableDiffusionXLInpaintPipeline(
        vae=self.vae,
        text_encoder=self.text,
        text_encoder_2=self.text2,
        unet=self.unet,
        scheduler=self.scheduler,
        tokenizer=None,
        tokenizer_2=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)
    self.num_channels_unet = self.unet.config.in_channels

ControlNetForText2ImageGeneration¤

Bases: GenericStableModel

Source code in src/unitorch/models/diffusers/modeling_controlnet.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    vae_config_path: str,
    controlnet_configs_path: Union[str, List[str]],
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    freeze_unet_encoder: Optional[bool] = False,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        vae_config_path=vae_config_path,
        controlnet_configs_path=controlnet_configs_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        freeze_unet_encoder=freeze_unet_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )
    self.pipeline = StableDiffusionControlNetPipeline(
        vae=self.vae,
        text_encoder=self.text,
        unet=self.unet,
        controlnet=self.controlnet,
        scheduler=self.scheduler,
        tokenizer=None,
        safety_checker=None,
        feature_extractor=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

ControlNetForImage2ImageGeneration¤

Bases: GenericStableModel

Source code in src/unitorch/models/diffusers/modeling_controlnet.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    vae_config_path: str,
    controlnet_configs_path: Union[str, List[str]],
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    freeze_unet_encoder: Optional[bool] = False,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        vae_config_path=vae_config_path,
        controlnet_configs_path=controlnet_configs_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        freeze_unet_encoder=freeze_unet_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )
    self.pipeline = StableDiffusionControlNetImg2ImgPipeline(
        vae=self.vae,
        text_encoder=self.text,
        unet=self.unet,
        controlnet=self.controlnet,
        scheduler=self.scheduler,
        tokenizer=None,
        safety_checker=None,
        feature_extractor=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

ControlNetForImageInpainting¤

Bases: GenericStableModel

Source code in src/unitorch/models/diffusers/modeling_controlnet.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    vae_config_path: str,
    controlnet_configs_path: Union[str, List[str]],
    scheduler_config_path: str,
    inpainting_controlnet_config_path: Union[str] = None,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    freeze_unet_encoder: Optional[bool] = False,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        vae_config_path=vae_config_path,
        controlnet_configs_path=controlnet_configs_path,
        scheduler_config_path=scheduler_config_path,
        inpainting_controlnet_config_path=inpainting_controlnet_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        freeze_unet_encoder=freeze_unet_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )
    self.pipeline = StableDiffusionControlNetInpaintPipeline(
        vae=self.vae,
        text_encoder=self.text,
        unet=self.unet,
        controlnet=self.controlnet,
        scheduler=self.scheduler,
        tokenizer=None,
        safety_checker=None,
        feature_extractor=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

ControlNetXLForText2ImageGeneration¤

Bases: GenericStableXLModel

ControlNetXL model for text-to-image generation.

Parameters:

Name Type Description Default
config_path str

Path to the model configuration file.

 required
text_config_path str

Path to the text model configuration file.

 required
text2_config_path str

Path to the second text model configuration file.

 required
vae_config_path str

Path to the VAE model configuration file.

 required
controlnet_configs_path str

Path to the ControlNet model configuration file.

 required
scheduler_config_path str

Path to the scheduler configuration file.

 required
quant_config_path Optional[str]

Path to the quantization configuration file (default: None).

 None
image_size Optional[int]

Size of the input image (default: None).

 None
in_channels Optional[int]

Number of input channels (default: None).

 None
out_channels Optional[int]

Number of output channels (default: None).

 None
num_train_timesteps Optional[int]

Number of training timesteps (default: 1000).

 1000
num_infer_timesteps Optional[int]

Number of inference timesteps (default: 50).

 50
freeze_vae_encoder Optional[bool]

Whether to freeze the VAE encoder (default: True).

 True
freeze_text_encoder Optional[bool]

Whether to freeze the text encoder (default: True).

 True
freeze_unet_encoder Optional[bool]

Whether to freeze the UNet encoder (default: True).

 False
seed Optional[int]

Random seed (default: 1123).

 1123
Source code in src/unitorch/models/diffusers/modeling_controlnet_xl.py 
 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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    text2_config_path: str,
    vae_config_path: str,
    controlnet_configs_path: Union[str, List[str]],
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    freeze_unet_encoder: Optional[bool] = False,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        text2_config_path=text2_config_path,
        vae_config_path=vae_config_path,
        controlnet_configs_path=controlnet_configs_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        freeze_unet_encoder=freeze_unet_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )
    self.pipeline = StableDiffusionXLControlNetPipeline(
        vae=self.vae,
        text_encoder=self.text,
        text_encoder_2=self.text2,
        unet=self.unet,
        controlnet=self.controlnet,
        scheduler=self.scheduler,
        tokenizer=None,
        tokenizer_2=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

forward ¤

forward(
    input_ids: Tensor,
    input2_ids: Tensor,
    add_time_ids: Tensor,
    pixel_values: Tensor,
    condition_pixel_values: Tensor,
    attention_mask: Optional[Tensor] = None,
    attention2_mask: Optional[Tensor] = None,
)

Forward pass of the model.

Parameters:

Name Type Description Default
input_ids Tensor

Input IDs.

 required
input2_ids Tensor

Second input IDs.

 required
add_time_ids Tensor

Additional time IDs.

 required
pixel_values Tensor

Pixel values.

 required
condition_pixel_values Tensor

Condition pixel values.

 required
attention_mask Optional[Tensor]

Attention mask (default: None).

 None
attention2_mask Optional[Tensor]

Second attention mask (default: None).

 None

Returns:

Type Description

torch.Tensor: Loss value.
Source code in src/unitorch/models/diffusers/modeling_controlnet_xl.py 
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
def forward(
    self,
    input_ids: torch.Tensor,
    input2_ids: torch.Tensor,
    add_time_ids: torch.Tensor,
    pixel_values: torch.Tensor,
    condition_pixel_values: torch.Tensor,
    attention_mask: Optional[torch.Tensor] = None,
    attention2_mask: Optional[torch.Tensor] = None,
):
    """
    Forward pass of the model.

    Args:
        input_ids (torch.Tensor): Input IDs.
        input2_ids (torch.Tensor): Second input IDs.
        add_time_ids (torch.Tensor): Additional time IDs.
        pixel_values (torch.Tensor): Pixel values.
        condition_pixel_values (torch.Tensor): Condition pixel values.
        attention_mask (Optional[torch.Tensor]): Attention mask (default: None).
        attention2_mask (Optional[torch.Tensor]): Second attention mask (default: None).

    Returns:
        torch.Tensor: Loss value.
    """
    prompt_outputs = self.text(
        input_ids,
        # attention_mask,
        output_hidden_states=True,
    )
    prompt_embeds = prompt_outputs.hidden_states[-2]
    prompt2_outputs = self.text2(
        input2_ids,
        # attention2_mask,
        output_hidden_states=True,
    )
    prompt2_embeds = prompt2_outputs.hidden_states[-2]
    prompt_embeds = torch.concat([prompt_embeds, prompt2_embeds], dim=-1)
    pooled_prompt_embeds = prompt2_outputs[0]

    latents = self.vae.encode(pixel_values).latent_dist.sample()
    latents = latents * self.vae.config.scaling_factor

    noise = torch.randn(latents.shape).to(latents.device)
    batch = latents.size(0)

    timesteps = torch.randint(
        0,
        self.scheduler.config.num_train_timesteps,
        (batch,),
        device=pixel_values.device,
    ).long()

    noise_latents = self.scheduler.add_noise(
        latents,
        noise,
        timesteps,
    )

    down_block_res_samples, mid_block_res_sample = self.controlnet(
        noise_latents,
        timesteps,
        encoder_hidden_states=prompt_embeds,
        controlnet_cond=condition_pixel_values,
        added_cond_kwargs={
            "time_ids": add_time_ids,
            "text_embeds": pooled_prompt_embeds,
        },
        return_dict=False,
    )
    outputs = self.unet(
        noise_latents,
        timesteps,
        prompt_embeds,
        added_cond_kwargs={
            "time_ids": add_time_ids,
            "text_embeds": pooled_prompt_embeds,
        },
        down_block_additional_residuals=down_block_res_samples,
        mid_block_additional_residual=mid_block_res_sample,
    ).sample

    if self.scheduler.config.prediction_type == "v_prediction":
        noise = self.scheduler.get_velocity(latents, noise, timesteps)

    loss = F.mse_loss(outputs, noise, reduction="mean")
    return loss

generate ¤

generate(
    condition_pixel_values: Tensor,
    input_ids: Tensor,
    input2_ids: Tensor,
    negative_input_ids: Tensor,
    negative_input2_ids: Tensor,
    attention_mask: Optional[Tensor] = None,
    attention2_mask: Optional[Tensor] = None,
    negative_attention_mask: Optional[Tensor] = None,
    negative_attention2_mask: Optional[Tensor] = None,
    height: Optional[int] = 1024,
    width: Optional[int] = 1024,
    guidance_scale: Optional[float] = 5.0,
    controlnet_conditioning_scale: Optional[
        Union[float, List[float]]
    ] = 1.0,
)

Generate images using the model.

Parameters:

Name Type Description Default
condition_pixel_values Tensor

Condition pixel values.

 required
input_ids Tensor

Input IDs.

 required
input2_ids Tensor

Second input IDs.

 required
negative_input_ids Tensor

Negative input IDs.

 required
negative_input2_ids Tensor

Negative second input IDs.

 required
attention_mask Optional[Tensor]

Attention mask (default: None).

 None
attention2_mask Optional[Tensor]

Second attention mask (default: None).

 None
negative_attention_mask Optional[Tensor]

Negative attention mask (default: None).

 None
negative_attention2_mask Optional[Tensor]

Negative second attention mask (default: None).

 None
height Optional[int]

Height of the generated images (default: 1024).

 1024
width Optional[int]

Width of the generated images (default: 1024).

 1024
guidance_scale Optional[float]

Scale for guidance (default: 5.0).

 5.0

Returns:

Name Type Description
GenericOutputs

Generated images.
Source code in src/unitorch/models/diffusers/modeling_controlnet_xl.py 
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
def generate(
    self,
    condition_pixel_values: torch.Tensor,
    input_ids: torch.Tensor,
    input2_ids: torch.Tensor,
    negative_input_ids: torch.Tensor,
    negative_input2_ids: torch.Tensor,
    attention_mask: Optional[torch.Tensor] = None,
    attention2_mask: Optional[torch.Tensor] = None,
    negative_attention_mask: Optional[torch.Tensor] = None,
    negative_attention2_mask: Optional[torch.Tensor] = None,
    height: Optional[int] = 1024,
    width: Optional[int] = 1024,
    guidance_scale: Optional[float] = 5.0,
    controlnet_conditioning_scale: Optional[Union[float, List[float]]] = 1.0,
):
    """
    Generate images using the model.

    Args:
        condition_pixel_values (torch.Tensor): Condition pixel values.
        input_ids (torch.Tensor): Input IDs.
        input2_ids (torch.Tensor): Second input IDs.
        negative_input_ids (torch.Tensor): Negative input IDs.
        negative_input2_ids (torch.Tensor): Negative second input IDs.
        attention_mask (Optional[torch.Tensor]): Attention mask (default: None).
        attention2_mask (Optional[torch.Tensor]): Second attention mask (default: None).
        negative_attention_mask (Optional[torch.Tensor]): Negative attention mask (default: None).
        negative_attention2_mask (Optional[torch.Tensor]): Negative second attention mask (default: None).
        height (Optional[int]): Height of the generated images (default: 1024).
        width (Optional[int]): Width of the generated images (default: 1024).
        guidance_scale (Optional[float]): Scale for guidance (default: 5.0).

    Returns:
        GenericOutputs: Generated images.
    """
    outputs = self.get_prompt_outputs(
        input_ids=input_ids,
        input2_ids=input2_ids,
        negative_input_ids=negative_input_ids,
        negative_input2_ids=negative_input2_ids,
        attention_mask=attention_mask,
        attention2_mask=attention2_mask,
        negative_attention_mask=negative_attention_mask,
        negative_attention2_mask=negative_attention2_mask,
    )
    if controlnet_conditioning_scale is None:
        if self.num_controlnets == 1:
            controlnet_conditioning_scale = 1.0
        else:
            controlnet_conditioning_scale = [1.0] * self.num_controlnets
    elif (
        not isinstance(controlnet_conditioning_scale, list)
        and self.num_controlnets > 1
    ):
        controlnet_conditioning_scale = [
            controlnet_conditioning_scale
        ] * self.num_controlnets

    images = self.pipeline(
        image=condition_pixel_values
        if self.num_controlnets == 1
        else list(condition_pixel_values.transpose(0, 1)),
        prompt_embeds=outputs.prompt_embeds,
        negative_prompt_embeds=outputs.negative_prompt_embeds,
        pooled_prompt_embeds=outputs.pooled_prompt_embeds,
        negative_pooled_prompt_embeds=outputs.negative_pooled_prompt_embeds,
        generator=torch.Generator(device=self.pipeline.device).manual_seed(
            self.seed
        ),
        num_inference_steps=self.num_infer_timesteps,
        height=height,
        width=width,
        guidance_scale=guidance_scale,
        controlnet_conditioning_scale=controlnet_conditioning_scale,
        output_type="np.array",
    ).images

    return GenericOutputs(images=torch.from_numpy(images))

ControlNetXLForImage2ImageGeneration¤

Bases: GenericStableXLModel

Source code in src/unitorch/models/diffusers/modeling_controlnet_xl.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    text2_config_path: str,
    vae_config_path: str,
    controlnet_configs_path: Union[str, List[str]],
    scheduler_config_path: str,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    freeze_unet_encoder: Optional[bool] = False,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        text2_config_path=text2_config_path,
        vae_config_path=vae_config_path,
        controlnet_configs_path=controlnet_configs_path,
        scheduler_config_path=scheduler_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        freeze_unet_encoder=freeze_unet_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )
    self.pipeline = StableDiffusionXLControlNetImg2ImgPipeline(
        vae=self.vae,
        text_encoder=self.text,
        text_encoder_2=self.text2,
        unet=self.unet,
        controlnet=self.controlnet,
        scheduler=self.scheduler,
        tokenizer=None,
        tokenizer_2=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)

ControlNetXLForImageInpainting¤

Bases: GenericStableXLModel

Source code in src/unitorch/models/diffusers/modeling_controlnet_xl.py 
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
def __init__(
    self,
    config_path: str,
    text_config_path: str,
    text2_config_path: str,
    vae_config_path: str,
    controlnet_configs_path: Union[str, List[str]],
    scheduler_config_path: str,
    inpainting_controlnet_config_path: Union[str] = None,
    quant_config_path: Optional[str] = None,
    image_size: Optional[int] = None,
    in_channels: Optional[int] = None,
    out_channels: Optional[int] = None,
    num_train_timesteps: Optional[int] = 1000,
    num_infer_timesteps: Optional[int] = 50,
    freeze_vae_encoder: Optional[bool] = True,
    freeze_text_encoder: Optional[bool] = True,
    freeze_unet_encoder: Optional[bool] = False,
    snr_gamma: Optional[float] = 5.0,
    seed: Optional[int] = 1123,
):
    super().__init__(
        config_path=config_path,
        text_config_path=text_config_path,
        text2_config_path=text2_config_path,
        vae_config_path=vae_config_path,
        controlnet_configs_path=controlnet_configs_path,
        scheduler_config_path=scheduler_config_path,
        inpainting_controlnet_config_path=inpainting_controlnet_config_path,
        quant_config_path=quant_config_path,
        image_size=image_size,
        in_channels=in_channels,
        out_channels=out_channels,
        num_train_timesteps=num_train_timesteps,
        num_infer_timesteps=num_infer_timesteps,
        freeze_vae_encoder=freeze_vae_encoder,
        freeze_text_encoder=freeze_text_encoder,
        freeze_unet_encoder=freeze_unet_encoder,
        snr_gamma=snr_gamma,
        seed=seed,
    )
    self.pipeline = StableDiffusionXLControlNetInpaintPipeline(
        vae=self.vae,
        text_encoder=self.text,
        text_encoder_2=self.text2,
        unet=self.unet,
        controlnet=self.controlnet,
        scheduler=self.scheduler,
        tokenizer=None,
        tokenizer_2=None,
    )
    self.pipeline.set_progress_bar_config(disable=True)