supported_models.py 21.5 KB
Newer Older
1
2
3
4
5
6
7
import torch
from . import model_base
from . import utils

from . import sd1_clip
from . import sd2_clip
from . import sdxl_clip
8
9
import comfy.text_encoders.sd3_clip
import comfy.text_encoders.sa_t5
10
import comfy.text_encoders.aura_t5
11
import comfy.text_encoders.hydit
12
13

from . import supported_models_base
14
from . import latent_formats
15

16
17
from . import diffusers_convert

18
19
20
21
22
23
class SD15(supported_models_base.BASE):
    unet_config = {
        "context_dim": 768,
        "model_channels": 320,
        "use_linear_in_transformer": False,
        "adm_in_channels": None,
comfyanonymous's avatar
comfyanonymous committed
24
        "use_temporal_attention": False,
25
26
27
28
29
30
31
    }

    unet_extra_config = {
        "num_heads": 8,
        "num_head_channels": -1,
    }

32
    latent_format = latent_formats.SD15
33
34
35
36
37
38
39
40
41
42
43
44
45

    def process_clip_state_dict(self, state_dict):
        k = list(state_dict.keys())
        for x in k:
            if x.startswith("cond_stage_model.transformer.") and not x.startswith("cond_stage_model.transformer.text_model."):
                y = x.replace("cond_stage_model.transformer.", "cond_stage_model.transformer.text_model.")
                state_dict[y] = state_dict.pop(x)

        if 'cond_stage_model.transformer.text_model.embeddings.position_ids' in state_dict:
            ids = state_dict['cond_stage_model.transformer.text_model.embeddings.position_ids']
            if ids.dtype == torch.float32:
                state_dict['cond_stage_model.transformer.text_model.embeddings.position_ids'] = ids.round()

46
        replace_prefix = {}
47
48
        replace_prefix["cond_stage_model."] = "clip_l."
        state_dict = utils.state_dict_prefix_replace(state_dict, replace_prefix, filter_keys=True)
49
50
        return state_dict

51
    def process_clip_state_dict_for_saving(self, state_dict):
52
53
54
55
56
        pop_keys = ["clip_l.transformer.text_projection.weight", "clip_l.logit_scale"]
        for p in pop_keys:
            if p in state_dict:
                state_dict.pop(p)

57
58
59
        replace_prefix = {"clip_l.": "cond_stage_model."}
        return utils.state_dict_prefix_replace(state_dict, replace_prefix)

60
    def clip_target(self, state_dict={}):
61
62
63
64
65
66
67
68
        return supported_models_base.ClipTarget(sd1_clip.SD1Tokenizer, sd1_clip.SD1ClipModel)

class SD20(supported_models_base.BASE):
    unet_config = {
        "context_dim": 1024,
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "adm_in_channels": None,
comfyanonymous's avatar
comfyanonymous committed
69
        "use_temporal_attention": False,
70
71
    }

72
73
74
75
76
77
    unet_extra_config = {
        "num_heads": -1,
        "num_head_channels": 64,
        "attn_precision": torch.float32,
    }

78
    latent_format = latent_formats.SD15
79

80
    def model_type(self, state_dict, prefix=""):
81
        if self.unet_config["in_channels"] == 4: #SD2.0 inpainting models are not v prediction
82
            k = "{}output_blocks.11.1.transformer_blocks.0.norm1.bias".format(prefix)
comfyanonymous's avatar
comfyanonymous committed
83
84
            out = state_dict.get(k, None)
            if out is not None and torch.std(out, unbiased=False) > 0.09: # not sure how well this will actually work. I guess we will find out.
85
86
                return model_base.ModelType.V_PREDICTION
        return model_base.ModelType.EPS
87
88

    def process_clip_state_dict(self, state_dict):
89
        replace_prefix = {}
90
91
92
        replace_prefix["conditioner.embedders.0.model."] = "clip_h." #SD2 in sgm format
        replace_prefix["cond_stage_model.model."] = "clip_h."
        state_dict = utils.state_dict_prefix_replace(state_dict, replace_prefix, filter_keys=True)
93
        state_dict = utils.clip_text_transformers_convert(state_dict, "clip_h.", "clip_h.transformer.")
94
95
        return state_dict

96
97
    def process_clip_state_dict_for_saving(self, state_dict):
        replace_prefix = {}
98
        replace_prefix["clip_h"] = "cond_stage_model.model"
99
        state_dict = utils.state_dict_prefix_replace(state_dict, replace_prefix)
100
101
102
        state_dict = diffusers_convert.convert_text_enc_state_dict_v20(state_dict)
        return state_dict

103
    def clip_target(self, state_dict={}):
104
105
106
107
108
109
110
111
        return supported_models_base.ClipTarget(sd2_clip.SD2Tokenizer, sd2_clip.SD2ClipModel)

class SD21UnclipL(SD20):
    unet_config = {
        "context_dim": 1024,
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "adm_in_channels": 1536,
comfyanonymous's avatar
comfyanonymous committed
112
        "use_temporal_attention": False,
113
114
115
116
117
118
119
120
121
122
123
124
    }

    clip_vision_prefix = "embedder.model.visual."
    noise_aug_config = {"noise_schedule_config": {"timesteps": 1000, "beta_schedule": "squaredcos_cap_v2"}, "timestep_dim": 768}


class SD21UnclipH(SD20):
    unet_config = {
        "context_dim": 1024,
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "adm_in_channels": 2048,
comfyanonymous's avatar
comfyanonymous committed
125
        "use_temporal_attention": False,
126
127
128
129
130
131
132
133
134
135
136
    }

    clip_vision_prefix = "embedder.model.visual."
    noise_aug_config = {"noise_schedule_config": {"timesteps": 1000, "beta_schedule": "squaredcos_cap_v2"}, "timestep_dim": 1024}

class SDXLRefiner(supported_models_base.BASE):
    unet_config = {
        "model_channels": 384,
        "use_linear_in_transformer": True,
        "context_dim": 1280,
        "adm_in_channels": 2560,
137
        "transformer_depth": [0, 0, 4, 4, 4, 4, 0, 0],
comfyanonymous's avatar
comfyanonymous committed
138
        "use_temporal_attention": False,
139
140
    }

141
    latent_format = latent_formats.SDXL
142

143
144
    def get_model(self, state_dict, prefix="", device=None):
        return model_base.SDXLRefiner(self, device=device)
145
146
147
148

    def process_clip_state_dict(self, state_dict):
        keys_to_replace = {}
        replace_prefix = {}
149
150
        replace_prefix["conditioner.embedders.0.model."] = "clip_g."
        state_dict = utils.state_dict_prefix_replace(state_dict, replace_prefix, filter_keys=True)
151

152
        state_dict = utils.clip_text_transformers_convert(state_dict, "clip_g.", "clip_g.transformer.")
153
        state_dict = utils.state_dict_key_replace(state_dict, keys_to_replace)
154
155
        return state_dict

156
157
158
    def process_clip_state_dict_for_saving(self, state_dict):
        replace_prefix = {}
        state_dict_g = diffusers_convert.convert_text_enc_state_dict_v20(state_dict, "clip_g")
159
160
        if "clip_g.transformer.text_model.embeddings.position_ids" in state_dict_g:
            state_dict_g.pop("clip_g.transformer.text_model.embeddings.position_ids")
161
        replace_prefix["clip_g"] = "conditioner.embedders.0.model"
162
        state_dict_g = utils.state_dict_prefix_replace(state_dict_g, replace_prefix)
163
164
        return state_dict_g

165
    def clip_target(self, state_dict={}):
166
167
168
169
170
171
        return supported_models_base.ClipTarget(sdxl_clip.SDXLTokenizer, sdxl_clip.SDXLRefinerClipModel)

class SDXL(supported_models_base.BASE):
    unet_config = {
        "model_channels": 320,
        "use_linear_in_transformer": True,
172
        "transformer_depth": [0, 0, 2, 2, 10, 10],
173
        "context_dim": 2048,
comfyanonymous's avatar
comfyanonymous committed
174
175
        "adm_in_channels": 2816,
        "use_temporal_attention": False,
176
177
    }

178
    latent_format = latent_formats.SDXL
179

180
    def model_type(self, state_dict, prefix=""):
181
182
183
184
185
186
        if 'edm_mean' in state_dict and 'edm_std' in state_dict: #Playground V2.5
            self.latent_format = latent_formats.SDXL_Playground_2_5()
            self.sampling_settings["sigma_data"] = 0.5
            self.sampling_settings["sigma_max"] = 80.0
            self.sampling_settings["sigma_min"] = 0.002
            return model_base.ModelType.EDM
comfyanonymous's avatar
comfyanonymous committed
187
188
189
190
191
        elif "edm_vpred.sigma_max" in state_dict:
            self.sampling_settings["sigma_max"] = float(state_dict["edm_vpred.sigma_max"].item())
            if "edm_vpred.sigma_min" in state_dict:
                self.sampling_settings["sigma_min"] = float(state_dict["edm_vpred.sigma_min"].item())
            return model_base.ModelType.V_PREDICTION_EDM
192
        elif "v_pred" in state_dict:
193
194
195
196
            return model_base.ModelType.V_PREDICTION
        else:
            return model_base.ModelType.EPS

197
    def get_model(self, state_dict, prefix="", device=None):
comfyanonymous's avatar
comfyanonymous committed
198
199
200
201
        out = model_base.SDXL(self, model_type=self.model_type(state_dict, prefix), device=device)
        if self.inpaint_model():
            out.set_inpaint()
        return out
202
203
204
205
206

    def process_clip_state_dict(self, state_dict):
        keys_to_replace = {}
        replace_prefix = {}

207
208
209
210
        replace_prefix["conditioner.embedders.0.transformer.text_model"] = "clip_l.transformer.text_model"
        replace_prefix["conditioner.embedders.1.model."] = "clip_g."
        state_dict = utils.state_dict_prefix_replace(state_dict, replace_prefix, filter_keys=True)

211
        state_dict = utils.state_dict_key_replace(state_dict, keys_to_replace)
212
        state_dict = utils.clip_text_transformers_convert(state_dict, "clip_g.", "clip_g.transformer.")
213
214
        return state_dict

215
216
217
218
219
220
221
222
    def process_clip_state_dict_for_saving(self, state_dict):
        replace_prefix = {}
        keys_to_replace = {}
        state_dict_g = diffusers_convert.convert_text_enc_state_dict_v20(state_dict, "clip_g")
        for k in state_dict:
            if k.startswith("clip_l"):
                state_dict_g[k] = state_dict[k]

223
224
225
226
227
228
        state_dict_g["clip_l.transformer.text_model.embeddings.position_ids"] = torch.arange(77).expand((1, -1))
        pop_keys = ["clip_l.transformer.text_projection.weight", "clip_l.logit_scale"]
        for p in pop_keys:
            if p in state_dict_g:
                state_dict_g.pop(p)

229
230
        replace_prefix["clip_g"] = "conditioner.embedders.1.model"
        replace_prefix["clip_l"] = "conditioner.embedders.0"
231
        state_dict_g = utils.state_dict_prefix_replace(state_dict_g, replace_prefix)
232
233
        return state_dict_g

234
    def clip_target(self, state_dict={}):
235
236
        return supported_models_base.ClipTarget(sdxl_clip.SDXLTokenizer, sdxl_clip.SDXLClipModel)

237
238
239
240
241
242
class SSD1B(SDXL):
    unet_config = {
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "transformer_depth": [0, 0, 2, 2, 4, 4],
        "context_dim": 2048,
comfyanonymous's avatar
comfyanonymous committed
243
244
        "adm_in_channels": 2816,
        "use_temporal_attention": False,
245
246
    }

comfyanonymous's avatar
comfyanonymous committed
247
248
249
250
251
252
253
254
255
256
class Segmind_Vega(SDXL):
    unet_config = {
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "transformer_depth": [0, 0, 1, 1, 2, 2],
        "context_dim": 2048,
        "adm_in_channels": 2816,
        "use_temporal_attention": False,
    }

comfyanonymous's avatar
comfyanonymous committed
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
class KOALA_700M(SDXL):
    unet_config = {
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "transformer_depth": [0, 2, 5],
        "context_dim": 2048,
        "adm_in_channels": 2816,
        "use_temporal_attention": False,
    }

class KOALA_1B(SDXL):
    unet_config = {
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "transformer_depth": [0, 2, 6],
        "context_dim": 2048,
        "adm_in_channels": 2816,
        "use_temporal_attention": False,
    }

comfyanonymous's avatar
comfyanonymous committed
277
278
279
280
281
282
283
284
285
286
287
288
class SVD_img2vid(supported_models_base.BASE):
    unet_config = {
        "model_channels": 320,
        "in_channels": 8,
        "use_linear_in_transformer": True,
        "transformer_depth": [1, 1, 1, 1, 1, 1, 0, 0],
        "context_dim": 1024,
        "adm_in_channels": 768,
        "use_temporal_attention": True,
        "use_temporal_resblock": True
    }

289
290
291
292
293
294
    unet_extra_config = {
        "num_heads": -1,
        "num_head_channels": 64,
        "attn_precision": torch.float32,
    }

comfyanonymous's avatar
comfyanonymous committed
295
296
297
298
299
300
301
302
303
304
    clip_vision_prefix = "conditioner.embedders.0.open_clip.model.visual."

    latent_format = latent_formats.SD15

    sampling_settings = {"sigma_max": 700.0, "sigma_min": 0.002}

    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.SVD_img2vid(self, device=device)
        return out

305
    def clip_target(self, state_dict={}):
comfyanonymous's avatar
comfyanonymous committed
306
        return None
307

comfyanonymous's avatar
comfyanonymous committed
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
class SV3D_u(SVD_img2vid):
    unet_config = {
        "model_channels": 320,
        "in_channels": 8,
        "use_linear_in_transformer": True,
        "transformer_depth": [1, 1, 1, 1, 1, 1, 0, 0],
        "context_dim": 1024,
        "adm_in_channels": 256,
        "use_temporal_attention": True,
        "use_temporal_resblock": True
    }

    vae_key_prefix = ["conditioner.embedders.1.encoder."]

    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.SV3D_u(self, device=device)
        return out

class SV3D_p(SV3D_u):
    unet_config = {
        "model_channels": 320,
        "in_channels": 8,
        "use_linear_in_transformer": True,
        "transformer_depth": [1, 1, 1, 1, 1, 1, 0, 0],
        "context_dim": 1024,
        "adm_in_channels": 1280,
        "use_temporal_attention": True,
        "use_temporal_resblock": True
    }


    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.SV3D_p(self, device=device)
        return out

343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
class Stable_Zero123(supported_models_base.BASE):
    unet_config = {
        "context_dim": 768,
        "model_channels": 320,
        "use_linear_in_transformer": False,
        "adm_in_channels": None,
        "use_temporal_attention": False,
        "in_channels": 8,
    }

    unet_extra_config = {
        "num_heads": 8,
        "num_head_channels": -1,
    }

comfyanonymous's avatar
comfyanonymous committed
358
359
360
361
362
    required_keys = {
        "cc_projection.weight": None,
        "cc_projection.bias": None,
    }

363
364
365
366
367
368
369
370
    clip_vision_prefix = "cond_stage_model.model.visual."

    latent_format = latent_formats.SD15

    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.Stable_Zero123(self, device=device, cc_projection_weight=state_dict["cc_projection.weight"], cc_projection_bias=state_dict["cc_projection.bias"])
        return out

371
    def clip_target(self, state_dict={}):
372
373
        return None

374
375
376
377
378
379
380
381
382
383
384
385
class SD_X4Upscaler(SD20):
    unet_config = {
        "context_dim": 1024,
        "model_channels": 256,
        'in_channels': 7,
        "use_linear_in_transformer": True,
        "adm_in_channels": None,
        "use_temporal_attention": False,
    }

    unet_extra_config = {
        "disable_self_attentions": [True, True, True, False],
386
        "num_classes": 1000,
387
388
389
390
391
392
393
394
395
396
397
398
399
400
        "num_heads": 8,
        "num_head_channels": -1,
    }

    latent_format = latent_formats.SD_X4

    sampling_settings = {
        "linear_start": 0.0001,
        "linear_end": 0.02,
    }

    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.SD_X4Upscaler(self, device=device)
        return out
401

comfyanonymous's avatar
comfyanonymous committed
402
403
404
405
406
407
408
409
410
411
class Stable_Cascade_C(supported_models_base.BASE):
    unet_config = {
        "stable_cascade_stage": 'c',
    }

    unet_extra_config = {}

    latent_format = latent_formats.SC_Prior
    supported_inference_dtypes = [torch.bfloat16, torch.float32]

412
413
414
415
    sampling_settings = {
        "shift": 2.0,
    }

416
417
418
419
    vae_key_prefix = ["vae."]
    text_encoder_key_prefix = ["text_encoder."]
    clip_vision_prefix = "clip_l_vision."

comfyanonymous's avatar
comfyanonymous committed
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
    def process_unet_state_dict(self, state_dict):
        key_list = list(state_dict.keys())
        for y in ["weight", "bias"]:
            suffix = "in_proj_{}".format(y)
            keys = filter(lambda a: a.endswith(suffix), key_list)
            for k_from in keys:
                weights = state_dict.pop(k_from)
                prefix = k_from[:-(len(suffix) + 1)]
                shape_from = weights.shape[0] // 3
                for x in range(3):
                    p = ["to_q", "to_k", "to_v"]
                    k_to = "{}.{}.{}".format(prefix, p[x], y)
                    state_dict[k_to] = weights[shape_from*x:shape_from*(x + 1)]
        return state_dict

435
436
437
438
439
440
    def process_clip_state_dict(self, state_dict):
        state_dict = utils.state_dict_prefix_replace(state_dict, {k: "" for k in self.text_encoder_key_prefix}, filter_keys=True)
        if "clip_g.text_projection" in state_dict:
            state_dict["clip_g.transformer.text_projection.weight"] = state_dict.pop("clip_g.text_projection").transpose(0, 1)
        return state_dict

comfyanonymous's avatar
comfyanonymous committed
441
442
443
444
    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.StableCascade_C(self, device=device)
        return out

445
    def clip_target(self, state_dict={}):
446
        return supported_models_base.ClipTarget(sdxl_clip.StableCascadeTokenizer, sdxl_clip.StableCascadeClipModel)
comfyanonymous's avatar
comfyanonymous committed
447

comfyanonymous's avatar
comfyanonymous committed
448
449
450
451
452
453
454
455
456
457
class Stable_Cascade_B(Stable_Cascade_C):
    unet_config = {
        "stable_cascade_stage": 'b',
    }

    unet_extra_config = {}

    latent_format = latent_formats.SC_B
    supported_inference_dtypes = [torch.float16, torch.bfloat16, torch.float32]

458
459
460
461
    sampling_settings = {
        "shift": 1.0,
    }

462
463
    clip_vision_prefix = None

comfyanonymous's avatar
comfyanonymous committed
464
465
466
467
    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.StableCascade_B(self, device=device)
        return out

comfyanonymous's avatar
comfyanonymous committed
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
class SD15_instructpix2pix(SD15):
    unet_config = {
        "context_dim": 768,
        "model_channels": 320,
        "use_linear_in_transformer": False,
        "adm_in_channels": None,
        "use_temporal_attention": False,
        "in_channels": 8,
    }

    def get_model(self, state_dict, prefix="", device=None):
        return model_base.SD15_instructpix2pix(self, device=device)

class SDXL_instructpix2pix(SDXL):
    unet_config = {
        "model_channels": 320,
        "use_linear_in_transformer": True,
        "transformer_depth": [0, 0, 2, 2, 10, 10],
        "context_dim": 2048,
        "adm_in_channels": 2816,
        "use_temporal_attention": False,
        "in_channels": 8,
    }

    def get_model(self, state_dict, prefix="", device=None):
comfyanonymous's avatar
comfyanonymous committed
493
        return model_base.SDXL_instructpix2pix(self, model_type=self.model_type(state_dict, prefix), device=device)
comfyanonymous's avatar
comfyanonymous committed
494

comfyanonymous's avatar
comfyanonymous committed
495
496
497
498
499
500
501
502
503
504
505
506
class SD3(supported_models_base.BASE):
    unet_config = {
        "in_channels": 16,
        "pos_embed_scaling_factor": None,
    }

    sampling_settings = {
        "shift": 3.0,
    }

    unet_extra_config = {}
    latent_format = latent_formats.SD3
507
    text_encoder_key_prefix = ["text_encoders."]
comfyanonymous's avatar
comfyanonymous committed
508
509
510
511
512

    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.SD3(self, device=device)
        return out

513
514
515
516
    def clip_target(self, state_dict={}):
        clip_l = False
        clip_g = False
        t5 = False
517
        dtype_t5 = None
518
519
520
521
522
        pref = self.text_encoder_key_prefix[0]
        if "{}clip_l.transformer.text_model.final_layer_norm.weight".format(pref) in state_dict:
            clip_l = True
        if "{}clip_g.transformer.text_model.final_layer_norm.weight".format(pref) in state_dict:
            clip_g = True
523
524
        t5_key = "{}t5xxl.transformer.encoder.final_layer_norm.weight".format(pref)
        if t5_key in state_dict:
525
            t5 = True
526
            dtype_t5 = state_dict[t5_key].dtype
527

528
        return supported_models_base.ClipTarget(comfy.text_encoders.sd3_clip.SD3Tokenizer, comfy.text_encoders.sd3_clip.sd3_clip(clip_l=clip_l, clip_g=clip_g, t5=t5, dtype_t5=dtype_t5))
comfyanonymous's avatar
comfyanonymous committed
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
class StableAudio(supported_models_base.BASE):
    unet_config = {
        "audio_model": "dit1.0",
    }

    sampling_settings = {"sigma_max": 500.0, "sigma_min": 0.03}

    unet_extra_config = {}
    latent_format = latent_formats.StableAudio1

    text_encoder_key_prefix = ["text_encoders."]
    vae_key_prefix = ["pretransform.model."]

    def get_model(self, state_dict, prefix="", device=None):
        seconds_start_sd = utils.state_dict_prefix_replace(state_dict, {"conditioner.conditioners.seconds_start.": ""}, filter_keys=True)
        seconds_total_sd = utils.state_dict_prefix_replace(state_dict, {"conditioner.conditioners.seconds_total.": ""}, filter_keys=True)
        return model_base.StableAudio1(self, seconds_start_embedder_weights=seconds_start_sd, seconds_total_embedder_weights=seconds_total_sd, device=device)

    def process_unet_state_dict(self, state_dict):
        for k in list(state_dict.keys()):
            if k.endswith(".cross_attend_norm.beta") or k.endswith(".ff_norm.beta") or k.endswith(".pre_norm.beta"): #These weights are all zero
                state_dict.pop(k)
        return state_dict

554
555
556
557
    def process_unet_state_dict_for_saving(self, state_dict):
        replace_prefix = {"": "model.model."}
        return utils.state_dict_prefix_replace(state_dict, replace_prefix)

558
    def clip_target(self, state_dict={}):
559
        return supported_models_base.ClipTarget(comfy.text_encoders.sa_t5.SAT5Tokenizer, comfy.text_encoders.sa_t5.SAT5Model)
560

561
562
563
564
565
566
567
class AuraFlow(supported_models_base.BASE):
    unet_config = {
        "cond_seq_dim": 2048,
    }

    sampling_settings = {
        "multiplier": 1.0,
568
        "shift": 1.73,
569
570
571
572
573
574
575
576
577
578
579
580
581
582
    }

    unet_extra_config = {}
    latent_format = latent_formats.SDXL

    vae_key_prefix = ["vae."]
    text_encoder_key_prefix = ["text_encoders."]

    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.AuraFlow(self, device=device)
        return out

    def clip_target(self, state_dict={}):
        return supported_models_base.ClipTarget(comfy.text_encoders.aura_t5.AuraT5Tokenizer, comfy.text_encoders.aura_t5.AuraT5Model)
comfyanonymous's avatar
comfyanonymous committed
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
class HunyuanDiT(supported_models_base.BASE):
    unet_config = {
        "image_model": "hydit",
    }

    unet_extra_config = {
        "attn_precision": torch.float32,
    }

    sampling_settings = {
        "linear_start": 0.00085,
        "linear_end": 0.018,
    }

    latent_format = latent_formats.SDXL

    vae_key_prefix = ["vae."]
    text_encoder_key_prefix = ["text_encoders."]

    def get_model(self, state_dict, prefix="", device=None):
        out = model_base.HunyuanDiT(self, device=device)
        return out

    def clip_target(self, state_dict={}):
        return supported_models_base.ClipTarget(comfy.text_encoders.hydit.HyditTokenizer, comfy.text_encoders.hydit.HyditModel)

class HunyuanDiT1(HunyuanDiT):
    unet_config = {
        "image_model": "hydit1",
    }

    unet_extra_config = {}

    sampling_settings = {
        "linear_start" : 0.00085,
        "linear_end" : 0.03,
    }


models = [Stable_Zero123, SD15_instructpix2pix, SD15, SD20, SD21UnclipL, SD21UnclipH, SDXL_instructpix2pix, SDXLRefiner, SDXL, SSD1B, KOALA_700M, KOALA_1B, Segmind_Vega, SD_X4Upscaler, Stable_Cascade_C, Stable_Cascade_B, SV3D_u, SV3D_p, SD3, StableAudio, AuraFlow, HunyuanDiT, HunyuanDiT1]
comfyanonymous's avatar
comfyanonymous committed
624

comfyanonymous's avatar
comfyanonymous committed
625
models += [SVD_img2vid]