auto_pipeline.py 52 KB
Newer Older
YiYi Xu's avatar
YiYi Xu committed
1
# coding=utf-8
2
# Copyright 2024 The HuggingFace Inc. team.
YiYi Xu's avatar
YiYi Xu committed
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
#
# 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.

from collections import OrderedDict

18
19
from huggingface_hub.utils import validate_hf_hub_args

YiYi Xu's avatar
YiYi Xu committed
20
from ..configuration_utils import ConfigMixin
21
from .aura_flow import AuraFlowPipeline
YiYi Xu's avatar
YiYi Xu committed
22
23
24
25
from .controlnet import (
    StableDiffusionControlNetImg2ImgPipeline,
    StableDiffusionControlNetInpaintPipeline,
    StableDiffusionControlNetPipeline,
26
    StableDiffusionXLControlNetImg2ImgPipeline,
27
    StableDiffusionXLControlNetInpaintPipeline,
YiYi Xu's avatar
YiYi Xu committed
28
29
30
    StableDiffusionXLControlNetPipeline,
)
from .deepfloyd_if import IFImg2ImgPipeline, IFInpaintingPipeline, IFPipeline
31
from .hunyuandit import HunyuanDiTPipeline
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
from .kandinsky import (
    KandinskyCombinedPipeline,
    KandinskyImg2ImgCombinedPipeline,
    KandinskyImg2ImgPipeline,
    KandinskyInpaintCombinedPipeline,
    KandinskyInpaintPipeline,
    KandinskyPipeline,
)
from .kandinsky2_2 import (
    KandinskyV22CombinedPipeline,
    KandinskyV22Img2ImgCombinedPipeline,
    KandinskyV22Img2ImgPipeline,
    KandinskyV22InpaintCombinedPipeline,
    KandinskyV22InpaintPipeline,
    KandinskyV22Pipeline,
)
48
from .kandinsky3 import Kandinsky3Img2ImgPipeline, Kandinsky3Pipeline
49
from .kolors import KolorsImg2ImgPipeline, KolorsPipeline
50
from .latent_consistency_models import LatentConsistencyModelImg2ImgPipeline, LatentConsistencyModelPipeline
YiYi Xu's avatar
YiYi Xu committed
51
from .pag import (
52
    StableDiffusionControlNetPAGPipeline,
53
    StableDiffusionPAGPipeline,
YiYi Xu's avatar
YiYi Xu committed
54
55
56
57
58
    StableDiffusionXLControlNetPAGPipeline,
    StableDiffusionXLPAGImg2ImgPipeline,
    StableDiffusionXLPAGInpaintPipeline,
    StableDiffusionXLPAGPipeline,
)
59
from .pixart_alpha import PixArtAlphaPipeline, PixArtSigmaPipeline
60
from .stable_cascade import StableCascadeCombinedPipeline, StableCascadeDecoderPipeline
YiYi Xu's avatar
YiYi Xu committed
61
62
63
64
65
from .stable_diffusion import (
    StableDiffusionImg2ImgPipeline,
    StableDiffusionInpaintPipeline,
    StableDiffusionPipeline,
)
66
67
from .stable_diffusion_3 import (
    StableDiffusion3Img2ImgPipeline,
68
    StableDiffusion3InpaintPipeline,
69
70
    StableDiffusion3Pipeline,
)
YiYi Xu's avatar
YiYi Xu committed
71
72
73
74
75
from .stable_diffusion_xl import (
    StableDiffusionXLImg2ImgPipeline,
    StableDiffusionXLInpaintPipeline,
    StableDiffusionXLPipeline,
)
Kashif Rasul's avatar
Kashif Rasul committed
76
from .wuerstchen import WuerstchenCombinedPipeline, WuerstchenDecoderPipeline
YiYi Xu's avatar
YiYi Xu committed
77
78
79
80
81
82


AUTO_TEXT2IMAGE_PIPELINES_MAPPING = OrderedDict(
    [
        ("stable-diffusion", StableDiffusionPipeline),
        ("stable-diffusion-xl", StableDiffusionXLPipeline),
83
        ("stable-diffusion-3", StableDiffusion3Pipeline),
YiYi Xu's avatar
YiYi Xu committed
84
        ("if", IFPipeline),
85
        ("hunyuan", HunyuanDiTPipeline),
86
87
        ("kandinsky", KandinskyCombinedPipeline),
        ("kandinsky22", KandinskyV22CombinedPipeline),
88
        ("kandinsky3", Kandinsky3Pipeline),
YiYi Xu's avatar
YiYi Xu committed
89
90
        ("stable-diffusion-controlnet", StableDiffusionControlNetPipeline),
        ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetPipeline),
Kashif Rasul's avatar
Kashif Rasul committed
91
        ("wuerstchen", WuerstchenCombinedPipeline),
92
        ("cascade", StableCascadeCombinedPipeline),
93
        ("lcm", LatentConsistencyModelPipeline),
94
95
        ("pixart-alpha", PixArtAlphaPipeline),
        ("pixart-sigma", PixArtSigmaPipeline),
96
        ("stable-diffusion-pag", StableDiffusionPAGPipeline),
97
        ("stable-diffusion-controlnet-pag", StableDiffusionControlNetPAGPipeline),
YiYi Xu's avatar
YiYi Xu committed
98
99
        ("stable-diffusion-xl-pag", StableDiffusionXLPAGPipeline),
        ("stable-diffusion-xl-controlnet-pag", StableDiffusionXLControlNetPAGPipeline),
100
101
        ("auraflow", AuraFlowPipeline),
        ("kolors", KolorsPipeline),
YiYi Xu's avatar
YiYi Xu committed
102
103
104
105
106
107
108
    ]
)

AUTO_IMAGE2IMAGE_PIPELINES_MAPPING = OrderedDict(
    [
        ("stable-diffusion", StableDiffusionImg2ImgPipeline),
        ("stable-diffusion-xl", StableDiffusionXLImg2ImgPipeline),
109
        ("stable-diffusion-3", StableDiffusion3Img2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
110
        ("if", IFImg2ImgPipeline),
111
112
        ("kandinsky", KandinskyImg2ImgCombinedPipeline),
        ("kandinsky22", KandinskyV22Img2ImgCombinedPipeline),
113
        ("kandinsky3", Kandinsky3Img2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
114
        ("stable-diffusion-controlnet", StableDiffusionControlNetImg2ImgPipeline),
115
        ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetImg2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
116
        ("stable-diffusion-xl-pag", StableDiffusionXLPAGImg2ImgPipeline),
117
        ("lcm", LatentConsistencyModelImg2ImgPipeline),
118
        ("kolors", KolorsImg2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
119
120
121
122
123
124
125
    ]
)

AUTO_INPAINT_PIPELINES_MAPPING = OrderedDict(
    [
        ("stable-diffusion", StableDiffusionInpaintPipeline),
        ("stable-diffusion-xl", StableDiffusionXLInpaintPipeline),
126
        ("stable-diffusion-3", StableDiffusion3InpaintPipeline),
YiYi Xu's avatar
YiYi Xu committed
127
        ("if", IFInpaintingPipeline),
128
129
130
        ("kandinsky", KandinskyInpaintCombinedPipeline),
        ("kandinsky22", KandinskyV22InpaintCombinedPipeline),
        ("stable-diffusion-controlnet", StableDiffusionControlNetInpaintPipeline),
131
        ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetInpaintPipeline),
YiYi Xu's avatar
YiYi Xu committed
132
        ("stable-diffusion-xl-pag", StableDiffusionXLPAGInpaintPipeline),
133
134
135
136
137
138
139
    ]
)

_AUTO_TEXT2IMAGE_DECODER_PIPELINES_MAPPING = OrderedDict(
    [
        ("kandinsky", KandinskyPipeline),
        ("kandinsky22", KandinskyV22Pipeline),
Kashif Rasul's avatar
Kashif Rasul committed
140
        ("wuerstchen", WuerstchenDecoderPipeline),
141
        ("cascade", StableCascadeDecoderPipeline),
142
143
144
145
146
147
148
149
150
151
    ]
)
_AUTO_IMAGE2IMAGE_DECODER_PIPELINES_MAPPING = OrderedDict(
    [
        ("kandinsky", KandinskyImg2ImgPipeline),
        ("kandinsky22", KandinskyV22Img2ImgPipeline),
    ]
)
_AUTO_INPAINT_DECODER_PIPELINES_MAPPING = OrderedDict(
    [
YiYi Xu's avatar
YiYi Xu committed
152
153
154
155
156
157
158
159
160
        ("kandinsky", KandinskyInpaintPipeline),
        ("kandinsky22", KandinskyV22InpaintPipeline),
    ]
)

SUPPORTED_TASKS_MAPPINGS = [
    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
    AUTO_INPAINT_PIPELINES_MAPPING,
161
162
163
    _AUTO_TEXT2IMAGE_DECODER_PIPELINES_MAPPING,
    _AUTO_IMAGE2IMAGE_DECODER_PIPELINES_MAPPING,
    _AUTO_INPAINT_DECODER_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
164
165
166
]


167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
def _get_connected_pipeline(pipeline_cls):
    # for now connected pipelines can only be loaded from decoder pipelines, such as kandinsky-community/kandinsky-2-2-decoder
    if pipeline_cls in _AUTO_TEXT2IMAGE_DECODER_PIPELINES_MAPPING.values():
        return _get_task_class(
            AUTO_TEXT2IMAGE_PIPELINES_MAPPING, pipeline_cls.__name__, throw_error_if_not_exist=False
        )
    if pipeline_cls in _AUTO_IMAGE2IMAGE_DECODER_PIPELINES_MAPPING.values():
        return _get_task_class(
            AUTO_IMAGE2IMAGE_PIPELINES_MAPPING, pipeline_cls.__name__, throw_error_if_not_exist=False
        )
    if pipeline_cls in _AUTO_INPAINT_DECODER_PIPELINES_MAPPING.values():
        return _get_task_class(AUTO_INPAINT_PIPELINES_MAPPING, pipeline_cls.__name__, throw_error_if_not_exist=False)


def _get_task_class(mapping, pipeline_class_name, throw_error_if_not_exist: bool = True):
YiYi Xu's avatar
YiYi Xu committed
182
183
184
185
186
187
188
189
190
191
192
193
    def get_model(pipeline_class_name):
        for task_mapping in SUPPORTED_TASKS_MAPPINGS:
            for model_name, pipeline in task_mapping.items():
                if pipeline.__name__ == pipeline_class_name:
                    return model_name

    model_name = get_model(pipeline_class_name)

    if model_name is not None:
        task_class = mapping.get(model_name, None)
        if task_class is not None:
            return task_class
194
195
196

    if throw_error_if_not_exist:
        raise ValueError(f"AutoPipeline can't find a pipeline linked to {pipeline_class_name} for {model_name}")
YiYi Xu's avatar
YiYi Xu committed
197
198
199
200
201


class AutoPipelineForText2Image(ConfigMixin):
    r"""

202
203
204
    [`AutoPipelineForText2Image`] is a generic pipeline class that instantiates a text-to-image pipeline class. The
    specific underlying pipeline class is automatically selected from either the
    [`~AutoPipelineForText2Image.from_pretrained`] or [`~AutoPipelineForText2Image.from_pipe`] methods.
YiYi Xu's avatar
YiYi Xu committed
205

206
    This class cannot be instantiated using `__init__()` (throws an error).
YiYi Xu's avatar
YiYi Xu committed
207
208
209
210
211
212
213

    Class attributes:

        - **config_name** (`str`) -- The configuration filename that stores the class and module names of all the
          diffusion pipeline's components.

    """
214

YiYi Xu's avatar
YiYi Xu committed
215
216
217
218
219
220
221
222
223
224
    config_name = "model_index.json"

    def __init__(self, *args, **kwargs):
        raise EnvironmentError(
            f"{self.__class__.__name__} is designed to be instantiated "
            f"using the `{self.__class__.__name__}.from_pretrained(pretrained_model_name_or_path)` or "
            f"`{self.__class__.__name__}.from_pipe(pipeline)` methods."
        )

    @classmethod
225
    @validate_hf_hub_args
YiYi Xu's avatar
YiYi Xu committed
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
    def from_pretrained(cls, pretrained_model_or_path, **kwargs):
        r"""
        Instantiates a text-to-image Pytorch diffusion pipeline from pretrained pipeline weight.

        The from_pretrained() method takes care of returning the correct pipeline class instance by:
            1. Detect the pipeline class of the pretrained_model_or_path based on the _class_name property of its
               config object
            2. Find the text-to-image pipeline linked to the pipeline class using pattern matching on pipeline class
               name.

        If a `controlnet` argument is passed, it will instantiate a [`StableDiffusionControlNetPipeline`] object.

        The pipeline is set in evaluation mode (`model.eval()`) by default.

        If you get the error message below, you need to finetune the weights for your downstream task:

        ```
        Some weights of UNet2DConditionModel were not initialized from the model checkpoint at runwayml/stable-diffusion-v1-5 and are newly initialized because the shapes did not match:
        - conv_in.weight: found shape torch.Size([320, 4, 3, 3]) in the checkpoint and torch.Size([320, 9, 3, 3]) in the model instantiated
        You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
        ```

        Parameters:
249
            pretrained_model_or_path (`str` or `os.PathLike`, *optional*):
YiYi Xu's avatar
YiYi Xu committed
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
                Can be either:

                    - A string, the *repo id* (for example `CompVis/ldm-text2im-large-256`) of a pretrained pipeline
                      hosted on the Hub.
                    - A path to a *directory* (for example `./my_pipeline_directory/`) containing pipeline weights
                      saved using
                    [`~DiffusionPipeline.save_pretrained`].
            torch_dtype (`str` or `torch.dtype`, *optional*):
                Override the default `torch.dtype` and load the model with another dtype. If "auto" is passed, the
                dtype is automatically derived from the model's weights.
            force_download (`bool`, *optional*, defaults to `False`):
                Whether or not to force the (re-)download of the model weights and configuration files, overriding the
                cached versions if they exist.
            cache_dir (`Union[str, os.PathLike]`, *optional*):
                Path to a directory where a downloaded pretrained model configuration is cached if the standard cache
                is not used.
266

YiYi Xu's avatar
YiYi Xu committed
267
268
269
270
271
272
273
274
            proxies (`Dict[str, str]`, *optional*):
                A dictionary of proxy servers to use by protocol or endpoint, for example, `{'http': 'foo.bar:3128',
                'http://hostname': 'foo.bar:4012'}`. The proxies are used on each request.
            output_loading_info(`bool`, *optional*, defaults to `False`):
                Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages.
            local_files_only (`bool`, *optional*, defaults to `False`):
                Whether to only load local model weights and configuration files or not. If set to `True`, the model
                won't be downloaded from the Hub.
275
            token (`str` or *bool*, *optional*):
YiYi Xu's avatar
YiYi Xu committed
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
                The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                `diffusers-cli login` (stored in `~/.huggingface`) is used.
            revision (`str`, *optional*, defaults to `"main"`):
                The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                allowed by Git.
            custom_revision (`str`, *optional*, defaults to `"main"`):
                The specific model version to use. It can be a branch name, a tag name, or a commit id similar to
                `revision` when loading a custom pipeline from the Hub. It can be a 🤗 Diffusers version when loading a
                custom pipeline from GitHub, otherwise it defaults to `"main"` when loading from the Hub.
            mirror (`str`, *optional*):
                Mirror source to resolve accessibility issues if you’re downloading a model in China. We do not
                guarantee the timeliness or safety of the source, and you should refer to the mirror site for more
                information.
            device_map (`str` or `Dict[str, Union[int, str, torch.device]]`, *optional*):
                A map that specifies where each submodule should go. It doesn’t need to be defined for each
                parameter/buffer name; once a given module name is inside, every submodule of it will be sent to the
                same device.

                Set `device_map="auto"` to have 🤗 Accelerate automatically compute the most optimized `device_map`. For
                more information about each option see [designing a device
                map](https://hf.co/docs/accelerate/main/en/usage_guides/big_modeling#designing-a-device-map).
            max_memory (`Dict`, *optional*):
                A dictionary device identifier for the maximum memory. Will default to the maximum memory available for
                each GPU and the available CPU RAM if unset.
            offload_folder (`str` or `os.PathLike`, *optional*):
                The path to offload weights if device_map contains the value `"disk"`.
            offload_state_dict (`bool`, *optional*):
                If `True`, temporarily offloads the CPU state dict to the hard drive to avoid running out of CPU RAM if
                the weight of the CPU state dict + the biggest shard of the checkpoint does not fit. Defaults to `True`
                when there is some disk offload.
            low_cpu_mem_usage (`bool`, *optional*, defaults to `True` if torch version >= 1.9.0 else `False`):
                Speed up model loading only loading the pretrained weights and not initializing the weights. This also
                tries to not use more than 1x model size in CPU memory (including peak memory) while loading the model.
                Only supported for PyTorch >= 1.9.0. If you are using an older version of PyTorch, setting this
                argument to `True` will raise an error.
            use_safetensors (`bool`, *optional*, defaults to `None`):
                If set to `None`, the safetensors weights are downloaded if they're available **and** if the
                safetensors library is installed. If set to `True`, the model is forcibly loaded from safetensors
                weights. If set to `False`, safetensors weights are not loaded.
            kwargs (remaining dictionary of keyword arguments, *optional*):
                Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline
                class). The overwritten components are passed directly to the pipelines `__init__` method. See example
                below for more information.
            variant (`str`, *optional*):
                Load weights from a specified variant filename such as `"fp16"` or `"ema"`. This is ignored when
                loading `from_flax`.

        <Tip>

        To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with
        `huggingface-cli login`.

        </Tip>

        Examples:

        ```py
333
        >>> from diffusers import AutoPipelineForText2Image
YiYi Xu's avatar
YiYi Xu committed
334

335
        >>> pipeline = AutoPipelineForText2Image.from_pretrained("runwayml/stable-diffusion-v1-5")
336
        >>> image = pipeline(prompt).images[0]
YiYi Xu's avatar
YiYi Xu committed
337
338
        ```
        """
339
        cache_dir = kwargs.pop("cache_dir", None)
340
341
        force_download = kwargs.pop("force_download", False)
        proxies = kwargs.pop("proxies", None)
342
        token = kwargs.pop("token", None)
343
344
345
346
347
348
349
        local_files_only = kwargs.pop("local_files_only", False)
        revision = kwargs.pop("revision", None)

        load_config_kwargs = {
            "cache_dir": cache_dir,
            "force_download": force_download,
            "proxies": proxies,
350
            "token": token,
351
352
353
354
355
            "local_files_only": local_files_only,
            "revision": revision,
        }

        config = cls.load_config(pretrained_model_or_path, **load_config_kwargs)
YiYi Xu's avatar
YiYi Xu committed
356
357
358
359
        orig_class_name = config["_class_name"]

        if "controlnet" in kwargs:
            orig_class_name = config["_class_name"].replace("Pipeline", "ControlNetPipeline")
YiYi Xu's avatar
YiYi Xu committed
360
361
362
363
        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
                orig_class_name = orig_class_name.replace("Pipeline", "PAGPipeline")
YiYi Xu's avatar
YiYi Xu committed
364
365
366

        text_2_image_cls = _get_task_class(AUTO_TEXT2IMAGE_PIPELINES_MAPPING, orig_class_name)

367
        kwargs = {**load_config_kwargs, **kwargs}
YiYi Xu's avatar
YiYi Xu committed
368
369
370
371
372
373
374
375
376
377
378
        return text_2_image_cls.from_pretrained(pretrained_model_or_path, **kwargs)

    @classmethod
    def from_pipe(cls, pipeline, **kwargs):
        r"""
        Instantiates a text-to-image Pytorch diffusion pipeline from another instantiated diffusion pipeline class.

        The from_pipe() method takes care of returning the correct pipeline class instance by finding the text-to-image
        pipeline linked to the pipeline class using pattern matching on pipeline class name.

        All the modules the pipeline contains will be used to initialize the new pipeline without reallocating
M. Tolga Cangöz's avatar
M. Tolga Cangöz committed
379
        additional memory.
YiYi Xu's avatar
YiYi Xu committed
380
381
382
383
384
385
386
387

        The pipeline is set in evaluation mode (`model.eval()`) by default.

        Parameters:
            pipeline (`DiffusionPipeline`):
                an instantiated `DiffusionPipeline` object

        ```py
388
        >>> from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
YiYi Xu's avatar
YiYi Xu committed
389
390
391
392
393

        >>> pipe_i2i = AutoPipelineForImage2Image.from_pretrained(
        ...     "runwayml/stable-diffusion-v1-5", requires_safety_checker=False
        ... )

394
395
        >>> pipe_t2i = AutoPipelineForText2Image.from_pipe(pipe_i2i)
        >>> image = pipe_t2i(prompt).images[0]
YiYi Xu's avatar
YiYi Xu committed
396
397
398
399
400
401
402
403
404
        ```
        """

        original_config = dict(pipeline.config)
        original_cls_name = pipeline.__class__.__name__

        # derive the pipeline class to instantiate
        text_2_image_cls = _get_task_class(AUTO_TEXT2IMAGE_PIPELINES_MAPPING, original_cls_name)

405
406
        if "controlnet" in kwargs:
            if kwargs["controlnet"] is not None:
YiYi Xu's avatar
YiYi Xu committed
407
                to_replace = "PAGPipeline" if "PAG" in text_2_image_cls.__name__ else "Pipeline"
408
409
                text_2_image_cls = _get_task_class(
                    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
410
                    text_2_image_cls.__name__.replace("ControlNet", "").replace(to_replace, "ControlNet" + to_replace),
411
412
413
414
                )
            else:
                text_2_image_cls = _get_task_class(
                    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
415
416
417
418
419
420
421
422
423
424
425
426
427
428
                    text_2_image_cls.__name__.replace("ControlNet", ""),
                )

        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
                text_2_image_cls = _get_task_class(
                    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
                    text_2_image_cls.__name__.replace("PAG", "").replace("Pipeline", "PAGPipeline"),
                )
            else:
                text_2_image_cls = _get_task_class(
                    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
                    text_2_image_cls.__name__.replace("PAG", ""),
429
430
                )

YiYi Xu's avatar
YiYi Xu committed
431
        # define expected module and optional kwargs given the pipeline signature
432
        expected_modules, optional_kwargs = text_2_image_cls._get_signature_keys(text_2_image_cls)
YiYi Xu's avatar
YiYi Xu committed
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

        pretrained_model_name_or_path = original_config.pop("_name_or_path", None)

        # allow users pass modules in `kwargs` to override the original pipeline's components
        passed_class_obj = {k: kwargs.pop(k) for k in expected_modules if k in kwargs}
        original_class_obj = {
            k: pipeline.components[k]
            for k, v in pipeline.components.items()
            if k in expected_modules and k not in passed_class_obj
        }

        # allow users pass optional kwargs to override the original pipelines config attribute
        passed_pipe_kwargs = {k: kwargs.pop(k) for k in optional_kwargs if k in kwargs}
        original_pipe_kwargs = {
            k: original_config[k]
            for k, v in original_config.items()
            if k in optional_kwargs and k not in passed_pipe_kwargs
        }

        # config that were not expected by original pipeline is stored as private attribute
        # we will pass them as optional arguments if they can be accepted by the pipeline
        additional_pipe_kwargs = [
            k[1:]
            for k in original_config.keys()
            if k.startswith("_") and k[1:] in optional_kwargs and k[1:] not in passed_pipe_kwargs
        ]
        for k in additional_pipe_kwargs:
            original_pipe_kwargs[k] = original_config.pop(f"_{k}")

        text_2_image_kwargs = {**passed_class_obj, **original_class_obj, **passed_pipe_kwargs, **original_pipe_kwargs}

        # store unused config as private attribute
        unused_original_config = {
            f"{'' if k.startswith('_') else '_'}{k}": original_config[k]
            for k, v in original_config.items()
            if k not in text_2_image_kwargs
        }

        missing_modules = set(expected_modules) - set(pipeline._optional_components) - set(text_2_image_kwargs.keys())

        if len(missing_modules) > 0:
            raise ValueError(
475
                f"Pipeline {text_2_image_cls} expected {expected_modules}, but only {set(list(passed_class_obj.keys()) + list(original_class_obj.keys()))} were passed"
YiYi Xu's avatar
YiYi Xu committed
476
477
478
479
480
481
482
483
484
485
486
487
            )

        model = text_2_image_cls(**text_2_image_kwargs)
        model.register_to_config(_name_or_path=pretrained_model_name_or_path)
        model.register_to_config(**unused_original_config)

        return model


class AutoPipelineForImage2Image(ConfigMixin):
    r"""

488
489
490
    [`AutoPipelineForImage2Image`] is a generic pipeline class that instantiates an image-to-image pipeline class. The
    specific underlying pipeline class is automatically selected from either the
    [`~AutoPipelineForImage2Image.from_pretrained`] or [`~AutoPipelineForImage2Image.from_pipe`] methods.
YiYi Xu's avatar
YiYi Xu committed
491

492
    This class cannot be instantiated using `__init__()` (throws an error).
YiYi Xu's avatar
YiYi Xu committed
493
494
495
496
497
498
499

    Class attributes:

        - **config_name** (`str`) -- The configuration filename that stores the class and module names of all the
          diffusion pipeline's components.

    """
500

YiYi Xu's avatar
YiYi Xu committed
501
502
503
504
505
506
507
508
509
510
    config_name = "model_index.json"

    def __init__(self, *args, **kwargs):
        raise EnvironmentError(
            f"{self.__class__.__name__} is designed to be instantiated "
            f"using the `{self.__class__.__name__}.from_pretrained(pretrained_model_name_or_path)` or "
            f"`{self.__class__.__name__}.from_pipe(pipeline)` methods."
        )

    @classmethod
511
    @validate_hf_hub_args
YiYi Xu's avatar
YiYi Xu committed
512
513
514
515
516
517
518
519
520
521
    def from_pretrained(cls, pretrained_model_or_path, **kwargs):
        r"""
        Instantiates a image-to-image Pytorch diffusion pipeline from pretrained pipeline weight.

        The from_pretrained() method takes care of returning the correct pipeline class instance by:
            1. Detect the pipeline class of the pretrained_model_or_path based on the _class_name property of its
               config object
            2. Find the image-to-image pipeline linked to the pipeline class using pattern matching on pipeline class
               name.

522
523
        If a `controlnet` argument is passed, it will instantiate a [`StableDiffusionControlNetImg2ImgPipeline`]
        object.
YiYi Xu's avatar
YiYi Xu committed
524
525
526
527
528
529
530
531
532
533
534
535

        The pipeline is set in evaluation mode (`model.eval()`) by default.

        If you get the error message below, you need to finetune the weights for your downstream task:

        ```
        Some weights of UNet2DConditionModel were not initialized from the model checkpoint at runwayml/stable-diffusion-v1-5 and are newly initialized because the shapes did not match:
        - conv_in.weight: found shape torch.Size([320, 4, 3, 3]) in the checkpoint and torch.Size([320, 9, 3, 3]) in the model instantiated
        You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
        ```

        Parameters:
536
            pretrained_model_or_path (`str` or `os.PathLike`, *optional*):
YiYi Xu's avatar
YiYi Xu committed
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
                Can be either:

                    - A string, the *repo id* (for example `CompVis/ldm-text2im-large-256`) of a pretrained pipeline
                      hosted on the Hub.
                    - A path to a *directory* (for example `./my_pipeline_directory/`) containing pipeline weights
                      saved using
                    [`~DiffusionPipeline.save_pretrained`].
            torch_dtype (`str` or `torch.dtype`, *optional*):
                Override the default `torch.dtype` and load the model with another dtype. If "auto" is passed, the
                dtype is automatically derived from the model's weights.
            force_download (`bool`, *optional*, defaults to `False`):
                Whether or not to force the (re-)download of the model weights and configuration files, overriding the
                cached versions if they exist.
            cache_dir (`Union[str, os.PathLike]`, *optional*):
                Path to a directory where a downloaded pretrained model configuration is cached if the standard cache
                is not used.
553

YiYi Xu's avatar
YiYi Xu committed
554
555
556
557
558
559
560
561
            proxies (`Dict[str, str]`, *optional*):
                A dictionary of proxy servers to use by protocol or endpoint, for example, `{'http': 'foo.bar:3128',
                'http://hostname': 'foo.bar:4012'}`. The proxies are used on each request.
            output_loading_info(`bool`, *optional*, defaults to `False`):
                Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages.
            local_files_only (`bool`, *optional*, defaults to `False`):
                Whether to only load local model weights and configuration files or not. If set to `True`, the model
                won't be downloaded from the Hub.
562
            token (`str` or *bool*, *optional*):
YiYi Xu's avatar
YiYi Xu committed
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
                The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                `diffusers-cli login` (stored in `~/.huggingface`) is used.
            revision (`str`, *optional*, defaults to `"main"`):
                The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                allowed by Git.
            custom_revision (`str`, *optional*, defaults to `"main"`):
                The specific model version to use. It can be a branch name, a tag name, or a commit id similar to
                `revision` when loading a custom pipeline from the Hub. It can be a 🤗 Diffusers version when loading a
                custom pipeline from GitHub, otherwise it defaults to `"main"` when loading from the Hub.
            mirror (`str`, *optional*):
                Mirror source to resolve accessibility issues if you’re downloading a model in China. We do not
                guarantee the timeliness or safety of the source, and you should refer to the mirror site for more
                information.
            device_map (`str` or `Dict[str, Union[int, str, torch.device]]`, *optional*):
                A map that specifies where each submodule should go. It doesn’t need to be defined for each
                parameter/buffer name; once a given module name is inside, every submodule of it will be sent to the
                same device.

                Set `device_map="auto"` to have 🤗 Accelerate automatically compute the most optimized `device_map`. For
                more information about each option see [designing a device
                map](https://hf.co/docs/accelerate/main/en/usage_guides/big_modeling#designing-a-device-map).
            max_memory (`Dict`, *optional*):
                A dictionary device identifier for the maximum memory. Will default to the maximum memory available for
                each GPU and the available CPU RAM if unset.
            offload_folder (`str` or `os.PathLike`, *optional*):
                The path to offload weights if device_map contains the value `"disk"`.
            offload_state_dict (`bool`, *optional*):
                If `True`, temporarily offloads the CPU state dict to the hard drive to avoid running out of CPU RAM if
                the weight of the CPU state dict + the biggest shard of the checkpoint does not fit. Defaults to `True`
                when there is some disk offload.
            low_cpu_mem_usage (`bool`, *optional*, defaults to `True` if torch version >= 1.9.0 else `False`):
                Speed up model loading only loading the pretrained weights and not initializing the weights. This also
                tries to not use more than 1x model size in CPU memory (including peak memory) while loading the model.
                Only supported for PyTorch >= 1.9.0. If you are using an older version of PyTorch, setting this
                argument to `True` will raise an error.
            use_safetensors (`bool`, *optional*, defaults to `None`):
                If set to `None`, the safetensors weights are downloaded if they're available **and** if the
                safetensors library is installed. If set to `True`, the model is forcibly loaded from safetensors
                weights. If set to `False`, safetensors weights are not loaded.
            kwargs (remaining dictionary of keyword arguments, *optional*):
                Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline
                class). The overwritten components are passed directly to the pipelines `__init__` method. See example
                below for more information.
            variant (`str`, *optional*):
                Load weights from a specified variant filename such as `"fp16"` or `"ema"`. This is ignored when
                loading `from_flax`.

        <Tip>

        To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with
        `huggingface-cli login`.

        </Tip>

        Examples:

        ```py
620
        >>> from diffusers import AutoPipelineForImage2Image
YiYi Xu's avatar
YiYi Xu committed
621

622
623
        >>> pipeline = AutoPipelineForImage2Image.from_pretrained("runwayml/stable-diffusion-v1-5")
        >>> image = pipeline(prompt, image).images[0]
YiYi Xu's avatar
YiYi Xu committed
624
625
        ```
        """
626
        cache_dir = kwargs.pop("cache_dir", None)
627
628
        force_download = kwargs.pop("force_download", False)
        proxies = kwargs.pop("proxies", None)
629
        token = kwargs.pop("token", None)
630
631
632
633
634
635
636
        local_files_only = kwargs.pop("local_files_only", False)
        revision = kwargs.pop("revision", None)

        load_config_kwargs = {
            "cache_dir": cache_dir,
            "force_download": force_download,
            "proxies": proxies,
637
            "token": token,
638
639
640
641
642
            "local_files_only": local_files_only,
            "revision": revision,
        }

        config = cls.load_config(pretrained_model_or_path, **load_config_kwargs)
YiYi Xu's avatar
YiYi Xu committed
643
644
645
646
        orig_class_name = config["_class_name"]

        if "controlnet" in kwargs:
            orig_class_name = config["_class_name"].replace("Pipeline", "ControlNetPipeline")
YiYi Xu's avatar
YiYi Xu committed
647
648
649
650
        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
                orig_class_name = orig_class_name.replace("Pipeline", "PAGPipeline")
YiYi Xu's avatar
YiYi Xu committed
651
652
653

        image_2_image_cls = _get_task_class(AUTO_IMAGE2IMAGE_PIPELINES_MAPPING, orig_class_name)

654
        kwargs = {**load_config_kwargs, **kwargs}
YiYi Xu's avatar
YiYi Xu committed
655
656
657
658
659
660
661
662
663
664
665
        return image_2_image_cls.from_pretrained(pretrained_model_or_path, **kwargs)

    @classmethod
    def from_pipe(cls, pipeline, **kwargs):
        r"""
        Instantiates a image-to-image Pytorch diffusion pipeline from another instantiated diffusion pipeline class.

        The from_pipe() method takes care of returning the correct pipeline class instance by finding the
        image-to-image pipeline linked to the pipeline class using pattern matching on pipeline class name.

        All the modules the pipeline contains will be used to initialize the new pipeline without reallocating
M. Tolga Cangöz's avatar
M. Tolga Cangöz committed
666
        additional memory.
YiYi Xu's avatar
YiYi Xu committed
667
668
669
670
671
672
673
674
675
676

        The pipeline is set in evaluation mode (`model.eval()`) by default.

        Parameters:
            pipeline (`DiffusionPipeline`):
                an instantiated `DiffusionPipeline` object

        Examples:

        ```py
677
        >>> from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
YiYi Xu's avatar
YiYi Xu committed
678
679
680
681
682

        >>> pipe_t2i = AutoPipelineForText2Image.from_pretrained(
        ...     "runwayml/stable-diffusion-v1-5", requires_safety_checker=False
        ... )

683
684
        >>> pipe_i2i = AutoPipelineForImage2Image.from_pipe(pipe_t2i)
        >>> image = pipe_i2i(prompt, image).images[0]
YiYi Xu's avatar
YiYi Xu committed
685
686
687
688
689
690
691
692
693
        ```
        """

        original_config = dict(pipeline.config)
        original_cls_name = pipeline.__class__.__name__

        # derive the pipeline class to instantiate
        image_2_image_cls = _get_task_class(AUTO_IMAGE2IMAGE_PIPELINES_MAPPING, original_cls_name)

694
695
        if "controlnet" in kwargs:
            if kwargs["controlnet"] is not None:
YiYi Xu's avatar
YiYi Xu committed
696
697
698
                to_replace = "Img2ImgPipeline"
                if "PAG" in image_2_image_cls.__name__:
                    to_replace = "PAG" + to_replace
699
700
                image_2_image_cls = _get_task_class(
                    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
701
                    image_2_image_cls.__name__.replace("ControlNet", "").replace(
YiYi Xu's avatar
YiYi Xu committed
702
                        to_replace, "ControlNet" + to_replace
703
                    ),
704
705
706
707
                )
            else:
                image_2_image_cls = _get_task_class(
                    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
708
709
710
711
712
713
714
715
716
717
718
719
720
721
                    image_2_image_cls.__name__.replace("ControlNet", ""),
                )

        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
                image_2_image_cls = _get_task_class(
                    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
                    image_2_image_cls.__name__.replace("PAG", "").replace("Img2ImgPipeline", "PAGImg2ImgPipeline"),
                )
            else:
                image_2_image_cls = _get_task_class(
                    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
                    image_2_image_cls.__name__.replace("PAG", ""),
722
723
                )

YiYi Xu's avatar
YiYi Xu committed
724
        # define expected module and optional kwargs given the pipeline signature
725
        expected_modules, optional_kwargs = image_2_image_cls._get_signature_keys(image_2_image_cls)
YiYi Xu's avatar
YiYi Xu committed
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

        pretrained_model_name_or_path = original_config.pop("_name_or_path", None)

        # allow users pass modules in `kwargs` to override the original pipeline's components
        passed_class_obj = {k: kwargs.pop(k) for k in expected_modules if k in kwargs}
        original_class_obj = {
            k: pipeline.components[k]
            for k, v in pipeline.components.items()
            if k in expected_modules and k not in passed_class_obj
        }

        # allow users pass optional kwargs to override the original pipelines config attribute
        passed_pipe_kwargs = {k: kwargs.pop(k) for k in optional_kwargs if k in kwargs}
        original_pipe_kwargs = {
            k: original_config[k]
            for k, v in original_config.items()
            if k in optional_kwargs and k not in passed_pipe_kwargs
        }

        # config attribute that were not expected by original pipeline is stored as its private attribute
        # we will pass them as optional arguments if they can be accepted by the pipeline
        additional_pipe_kwargs = [
            k[1:]
            for k in original_config.keys()
            if k.startswith("_") and k[1:] in optional_kwargs and k[1:] not in passed_pipe_kwargs
        ]
        for k in additional_pipe_kwargs:
            original_pipe_kwargs[k] = original_config.pop(f"_{k}")

        image_2_image_kwargs = {**passed_class_obj, **original_class_obj, **passed_pipe_kwargs, **original_pipe_kwargs}

        # store unused config as private attribute
        unused_original_config = {
            f"{'' if k.startswith('_') else '_'}{k}": original_config[k]
            for k, v in original_config.items()
            if k not in image_2_image_kwargs
        }

        missing_modules = set(expected_modules) - set(pipeline._optional_components) - set(image_2_image_kwargs.keys())

        if len(missing_modules) > 0:
            raise ValueError(
768
                f"Pipeline {image_2_image_cls} expected {expected_modules}, but only {set(list(passed_class_obj.keys()) + list(original_class_obj.keys()))} were passed"
YiYi Xu's avatar
YiYi Xu committed
769
770
771
772
773
774
775
776
777
778
779
780
            )

        model = image_2_image_cls(**image_2_image_kwargs)
        model.register_to_config(_name_or_path=pretrained_model_name_or_path)
        model.register_to_config(**unused_original_config)

        return model


class AutoPipelineForInpainting(ConfigMixin):
    r"""

781
782
783
    [`AutoPipelineForInpainting`] is a generic pipeline class that instantiates an inpainting pipeline class. The
    specific underlying pipeline class is automatically selected from either the
    [`~AutoPipelineForInpainting.from_pretrained`] or [`~AutoPipelineForInpainting.from_pipe`] methods.
YiYi Xu's avatar
YiYi Xu committed
784

785
    This class cannot be instantiated using `__init__()` (throws an error).
YiYi Xu's avatar
YiYi Xu committed
786
787
788
789
790
791
792

    Class attributes:

        - **config_name** (`str`) -- The configuration filename that stores the class and module names of all the
          diffusion pipeline's components.

    """
793

YiYi Xu's avatar
YiYi Xu committed
794
795
796
797
798
799
800
801
802
803
    config_name = "model_index.json"

    def __init__(self, *args, **kwargs):
        raise EnvironmentError(
            f"{self.__class__.__name__} is designed to be instantiated "
            f"using the `{self.__class__.__name__}.from_pretrained(pretrained_model_name_or_path)` or "
            f"`{self.__class__.__name__}.from_pipe(pipeline)` methods."
        )

    @classmethod
804
    @validate_hf_hub_args
YiYi Xu's avatar
YiYi Xu committed
805
806
807
808
809
810
811
812
813
    def from_pretrained(cls, pretrained_model_or_path, **kwargs):
        r"""
        Instantiates a inpainting Pytorch diffusion pipeline from pretrained pipeline weight.

        The from_pretrained() method takes care of returning the correct pipeline class instance by:
            1. Detect the pipeline class of the pretrained_model_or_path based on the _class_name property of its
               config object
            2. Find the inpainting pipeline linked to the pipeline class using pattern matching on pipeline class name.

814
815
        If a `controlnet` argument is passed, it will instantiate a [`StableDiffusionControlNetInpaintPipeline`]
        object.
YiYi Xu's avatar
YiYi Xu committed
816
817
818
819
820
821
822
823
824
825
826
827

        The pipeline is set in evaluation mode (`model.eval()`) by default.

        If you get the error message below, you need to finetune the weights for your downstream task:

        ```
        Some weights of UNet2DConditionModel were not initialized from the model checkpoint at runwayml/stable-diffusion-v1-5 and are newly initialized because the shapes did not match:
        - conv_in.weight: found shape torch.Size([320, 4, 3, 3]) in the checkpoint and torch.Size([320, 9, 3, 3]) in the model instantiated
        You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
        ```

        Parameters:
828
            pretrained_model_or_path (`str` or `os.PathLike`, *optional*):
YiYi Xu's avatar
YiYi Xu committed
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
                Can be either:

                    - A string, the *repo id* (for example `CompVis/ldm-text2im-large-256`) of a pretrained pipeline
                      hosted on the Hub.
                    - A path to a *directory* (for example `./my_pipeline_directory/`) containing pipeline weights
                      saved using
                    [`~DiffusionPipeline.save_pretrained`].
            torch_dtype (`str` or `torch.dtype`, *optional*):
                Override the default `torch.dtype` and load the model with another dtype. If "auto" is passed, the
                dtype is automatically derived from the model's weights.
            force_download (`bool`, *optional*, defaults to `False`):
                Whether or not to force the (re-)download of the model weights and configuration files, overriding the
                cached versions if they exist.
            cache_dir (`Union[str, os.PathLike]`, *optional*):
                Path to a directory where a downloaded pretrained model configuration is cached if the standard cache
                is not used.
845

YiYi Xu's avatar
YiYi Xu committed
846
847
848
849
850
851
852
853
            proxies (`Dict[str, str]`, *optional*):
                A dictionary of proxy servers to use by protocol or endpoint, for example, `{'http': 'foo.bar:3128',
                'http://hostname': 'foo.bar:4012'}`. The proxies are used on each request.
            output_loading_info(`bool`, *optional*, defaults to `False`):
                Whether or not to also return a dictionary containing missing keys, unexpected keys and error messages.
            local_files_only (`bool`, *optional*, defaults to `False`):
                Whether to only load local model weights and configuration files or not. If set to `True`, the model
                won't be downloaded from the Hub.
854
            token (`str` or *bool*, *optional*):
YiYi Xu's avatar
YiYi Xu committed
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
                The token to use as HTTP bearer authorization for remote files. If `True`, the token generated from
                `diffusers-cli login` (stored in `~/.huggingface`) is used.
            revision (`str`, *optional*, defaults to `"main"`):
                The specific model version to use. It can be a branch name, a tag name, a commit id, or any identifier
                allowed by Git.
            custom_revision (`str`, *optional*, defaults to `"main"`):
                The specific model version to use. It can be a branch name, a tag name, or a commit id similar to
                `revision` when loading a custom pipeline from the Hub. It can be a 🤗 Diffusers version when loading a
                custom pipeline from GitHub, otherwise it defaults to `"main"` when loading from the Hub.
            mirror (`str`, *optional*):
                Mirror source to resolve accessibility issues if you’re downloading a model in China. We do not
                guarantee the timeliness or safety of the source, and you should refer to the mirror site for more
                information.
            device_map (`str` or `Dict[str, Union[int, str, torch.device]]`, *optional*):
                A map that specifies where each submodule should go. It doesn’t need to be defined for each
                parameter/buffer name; once a given module name is inside, every submodule of it will be sent to the
                same device.

                Set `device_map="auto"` to have 🤗 Accelerate automatically compute the most optimized `device_map`. For
                more information about each option see [designing a device
                map](https://hf.co/docs/accelerate/main/en/usage_guides/big_modeling#designing-a-device-map).
            max_memory (`Dict`, *optional*):
                A dictionary device identifier for the maximum memory. Will default to the maximum memory available for
                each GPU and the available CPU RAM if unset.
            offload_folder (`str` or `os.PathLike`, *optional*):
                The path to offload weights if device_map contains the value `"disk"`.
            offload_state_dict (`bool`, *optional*):
                If `True`, temporarily offloads the CPU state dict to the hard drive to avoid running out of CPU RAM if
                the weight of the CPU state dict + the biggest shard of the checkpoint does not fit. Defaults to `True`
                when there is some disk offload.
            low_cpu_mem_usage (`bool`, *optional*, defaults to `True` if torch version >= 1.9.0 else `False`):
                Speed up model loading only loading the pretrained weights and not initializing the weights. This also
                tries to not use more than 1x model size in CPU memory (including peak memory) while loading the model.
                Only supported for PyTorch >= 1.9.0. If you are using an older version of PyTorch, setting this
                argument to `True` will raise an error.
            use_safetensors (`bool`, *optional*, defaults to `None`):
                If set to `None`, the safetensors weights are downloaded if they're available **and** if the
                safetensors library is installed. If set to `True`, the model is forcibly loaded from safetensors
                weights. If set to `False`, safetensors weights are not loaded.
            kwargs (remaining dictionary of keyword arguments, *optional*):
                Can be used to overwrite load and saveable variables (the pipeline components of the specific pipeline
                class). The overwritten components are passed directly to the pipelines `__init__` method. See example
                below for more information.
            variant (`str`, *optional*):
                Load weights from a specified variant filename such as `"fp16"` or `"ema"`. This is ignored when
                loading `from_flax`.

        <Tip>

        To use private or [gated](https://huggingface.co/docs/hub/models-gated#gated-models) models, log-in with
        `huggingface-cli login`.

        </Tip>

        Examples:

        ```py
912
        >>> from diffusers import AutoPipelineForInpainting
YiYi Xu's avatar
YiYi Xu committed
913

914
915
        >>> pipeline = AutoPipelineForInpainting.from_pretrained("runwayml/stable-diffusion-v1-5")
        >>> image = pipeline(prompt, image=init_image, mask_image=mask_image).images[0]
YiYi Xu's avatar
YiYi Xu committed
916
917
        ```
        """
918
        cache_dir = kwargs.pop("cache_dir", None)
919
920
        force_download = kwargs.pop("force_download", False)
        proxies = kwargs.pop("proxies", None)
921
        token = kwargs.pop("token", None)
922
923
924
925
926
927
928
        local_files_only = kwargs.pop("local_files_only", False)
        revision = kwargs.pop("revision", None)

        load_config_kwargs = {
            "cache_dir": cache_dir,
            "force_download": force_download,
            "proxies": proxies,
929
            "token": token,
930
931
932
933
934
            "local_files_only": local_files_only,
            "revision": revision,
        }

        config = cls.load_config(pretrained_model_or_path, **load_config_kwargs)
YiYi Xu's avatar
YiYi Xu committed
935
936
937
938
        orig_class_name = config["_class_name"]

        if "controlnet" in kwargs:
            orig_class_name = config["_class_name"].replace("Pipeline", "ControlNetPipeline")
YiYi Xu's avatar
YiYi Xu committed
939
940
941
942
        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
                orig_class_name = config["_class_name"].replace("Pipeline", "PAGPipeline")
YiYi Xu's avatar
YiYi Xu committed
943
944
945

        inpainting_cls = _get_task_class(AUTO_INPAINT_PIPELINES_MAPPING, orig_class_name)

946
        kwargs = {**load_config_kwargs, **kwargs}
YiYi Xu's avatar
YiYi Xu committed
947
948
949
950
951
952
953
954
955
956
957
        return inpainting_cls.from_pretrained(pretrained_model_or_path, **kwargs)

    @classmethod
    def from_pipe(cls, pipeline, **kwargs):
        r"""
        Instantiates a inpainting Pytorch diffusion pipeline from another instantiated diffusion pipeline class.

        The from_pipe() method takes care of returning the correct pipeline class instance by finding the inpainting
        pipeline linked to the pipeline class using pattern matching on pipeline class name.

        All the modules the pipeline class contain will be used to initialize the new pipeline without reallocating
M. Tolga Cangöz's avatar
M. Tolga Cangöz committed
958
        additional memory.
YiYi Xu's avatar
YiYi Xu committed
959
960
961
962
963
964
965
966
967
968

        The pipeline is set in evaluation mode (`model.eval()`) by default.

        Parameters:
            pipeline (`DiffusionPipeline`):
                an instantiated `DiffusionPipeline` object

        Examples:

        ```py
969
        >>> from diffusers import AutoPipelineForText2Image, AutoPipelineForInpainting
YiYi Xu's avatar
YiYi Xu committed
970
971
972
973
974
975

        >>> pipe_t2i = AutoPipelineForText2Image.from_pretrained(
        ...     "DeepFloyd/IF-I-XL-v1.0", requires_safety_checker=False
        ... )

        >>> pipe_inpaint = AutoPipelineForInpainting.from_pipe(pipe_t2i)
976
        >>> image = pipe_inpaint(prompt, image=init_image, mask_image=mask_image).images[0]
YiYi Xu's avatar
YiYi Xu committed
977
978
979
980
981
982
983
984
        ```
        """
        original_config = dict(pipeline.config)
        original_cls_name = pipeline.__class__.__name__

        # derive the pipeline class to instantiate
        inpainting_cls = _get_task_class(AUTO_INPAINT_PIPELINES_MAPPING, original_cls_name)

985
986
987
988
        if "controlnet" in kwargs:
            if kwargs["controlnet"] is not None:
                inpainting_cls = _get_task_class(
                    AUTO_INPAINT_PIPELINES_MAPPING,
989
990
991
                    inpainting_cls.__name__.replace("ControlNet", "").replace(
                        "InpaintPipeline", "ControlNetInpaintPipeline"
                    ),
992
993
994
995
996
997
998
                )
            else:
                inpainting_cls = _get_task_class(
                    AUTO_INPAINT_PIPELINES_MAPPING,
                    inpainting_cls.__name__.replace("ControlNetInpaintPipeline", "InpaintPipeline"),
                )

YiYi Xu's avatar
YiYi Xu committed
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
                inpainting_cls = _get_task_class(
                    AUTO_INPAINT_PIPELINES_MAPPING,
                    inpainting_cls.__name__.replace("PAG", "").replace("InpaintPipeline", "PAGInpaintPipeline"),
                )
            else:
                inpainting_cls = _get_task_class(
                    AUTO_INPAINT_PIPELINES_MAPPING,
                    inpainting_cls.__name__.replace("PAGInpaintPipeline", "InpaintPipeline"),
                )

YiYi Xu's avatar
YiYi Xu committed
1012
        # define expected module and optional kwargs given the pipeline signature
1013
        expected_modules, optional_kwargs = inpainting_cls._get_signature_keys(inpainting_cls)
YiYi Xu's avatar
YiYi Xu committed
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

        pretrained_model_name_or_path = original_config.pop("_name_or_path", None)

        # allow users pass modules in `kwargs` to override the original pipeline's components
        passed_class_obj = {k: kwargs.pop(k) for k in expected_modules if k in kwargs}
        original_class_obj = {
            k: pipeline.components[k]
            for k, v in pipeline.components.items()
            if k in expected_modules and k not in passed_class_obj
        }

        # allow users pass optional kwargs to override the original pipelines config attribute
        passed_pipe_kwargs = {k: kwargs.pop(k) for k in optional_kwargs if k in kwargs}
        original_pipe_kwargs = {
            k: original_config[k]
            for k, v in original_config.items()
            if k in optional_kwargs and k not in passed_pipe_kwargs
        }

        # config that were not expected by original pipeline is stored as private attribute
        # we will pass them as optional arguments if they can be accepted by the pipeline
        additional_pipe_kwargs = [
            k[1:]
            for k in original_config.keys()
            if k.startswith("_") and k[1:] in optional_kwargs and k[1:] not in passed_pipe_kwargs
        ]
        for k in additional_pipe_kwargs:
            original_pipe_kwargs[k] = original_config.pop(f"_{k}")

        inpainting_kwargs = {**passed_class_obj, **original_class_obj, **passed_pipe_kwargs, **original_pipe_kwargs}

        # store unused config as private attribute
        unused_original_config = {
            f"{'' if k.startswith('_') else '_'}{k}": original_config[k]
            for k, v in original_config.items()
            if k not in inpainting_kwargs
        }

        missing_modules = set(expected_modules) - set(pipeline._optional_components) - set(inpainting_kwargs.keys())

        if len(missing_modules) > 0:
            raise ValueError(
1056
                f"Pipeline {inpainting_cls} expected {expected_modules}, but only {set(list(passed_class_obj.keys()) + list(original_class_obj.keys()))} were passed"
YiYi Xu's avatar
YiYi Xu committed
1057
1058
1059
1060
1061
1062
1063
            )

        model = inpainting_cls(**inpainting_kwargs)
        model.register_to_config(_name_or_path=pretrained_model_name_or_path)
        model.register_to_config(**unused_original_config)

        return model