auto_pipeline.py 54 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 ..utils import is_sentencepiece_available
22
from .aura_flow import AuraFlowPipeline
YiYi Xu's avatar
YiYi Xu committed
23
24
25
26
from .controlnet import (
    StableDiffusionControlNetImg2ImgPipeline,
    StableDiffusionControlNetInpaintPipeline,
    StableDiffusionControlNetPipeline,
27
    StableDiffusionXLControlNetImg2ImgPipeline,
28
    StableDiffusionXLControlNetInpaintPipeline,
YiYi Xu's avatar
YiYi Xu committed
29
30
31
    StableDiffusionXLControlNetPipeline,
)
from .deepfloyd_if import IFImg2ImgPipeline, IFInpaintingPipeline, IFPipeline
32
33
34
35
36
37
38
39
from .flux import (
    FluxControlNetImg2ImgPipeline,
    FluxControlNetInpaintPipeline,
    FluxControlNetPipeline,
    FluxImg2ImgPipeline,
    FluxInpaintPipeline,
    FluxPipeline,
)
40
from .hunyuandit import HunyuanDiTPipeline
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
from .kandinsky import (
    KandinskyCombinedPipeline,
    KandinskyImg2ImgCombinedPipeline,
    KandinskyImg2ImgPipeline,
    KandinskyInpaintCombinedPipeline,
    KandinskyInpaintPipeline,
    KandinskyPipeline,
)
from .kandinsky2_2 import (
    KandinskyV22CombinedPipeline,
    KandinskyV22Img2ImgCombinedPipeline,
    KandinskyV22Img2ImgPipeline,
    KandinskyV22InpaintCombinedPipeline,
    KandinskyV22InpaintPipeline,
    KandinskyV22Pipeline,
)
57
from .kandinsky3 import Kandinsky3Img2ImgPipeline, Kandinsky3Pipeline
58
from .latent_consistency_models import LatentConsistencyModelImg2ImgPipeline, LatentConsistencyModelPipeline
59
from .lumina import LuminaText2ImgPipeline
YiYi Xu's avatar
YiYi Xu committed
60
from .pag import (
61
    HunyuanDiTPAGPipeline,
62
    PixArtSigmaPAGPipeline,
63
    StableDiffusion3PAGPipeline,
64
    StableDiffusionControlNetPAGInpaintPipeline,
65
    StableDiffusionControlNetPAGPipeline,
66
    StableDiffusionPAGImg2ImgPipeline,
67
    StableDiffusionPAGPipeline,
68
    StableDiffusionXLControlNetPAGImg2ImgPipeline,
YiYi Xu's avatar
YiYi Xu committed
69
70
71
72
73
    StableDiffusionXLControlNetPAGPipeline,
    StableDiffusionXLPAGImg2ImgPipeline,
    StableDiffusionXLPAGInpaintPipeline,
    StableDiffusionXLPAGPipeline,
)
74
from .pixart_alpha import PixArtAlphaPipeline, PixArtSigmaPipeline
75
from .stable_cascade import StableCascadeCombinedPipeline, StableCascadeDecoderPipeline
YiYi Xu's avatar
YiYi Xu committed
76
77
78
79
80
from .stable_diffusion import (
    StableDiffusionImg2ImgPipeline,
    StableDiffusionInpaintPipeline,
    StableDiffusionPipeline,
)
81
82
from .stable_diffusion_3 import (
    StableDiffusion3Img2ImgPipeline,
83
    StableDiffusion3InpaintPipeline,
84
85
    StableDiffusion3Pipeline,
)
YiYi Xu's avatar
YiYi Xu committed
86
87
88
89
90
from .stable_diffusion_xl import (
    StableDiffusionXLImg2ImgPipeline,
    StableDiffusionXLInpaintPipeline,
    StableDiffusionXLPipeline,
)
Kashif Rasul's avatar
Kashif Rasul committed
91
from .wuerstchen import WuerstchenCombinedPipeline, WuerstchenDecoderPipeline
YiYi Xu's avatar
YiYi Xu committed
92
93
94
95
96
97


AUTO_TEXT2IMAGE_PIPELINES_MAPPING = OrderedDict(
    [
        ("stable-diffusion", StableDiffusionPipeline),
        ("stable-diffusion-xl", StableDiffusionXLPipeline),
98
        ("stable-diffusion-3", StableDiffusion3Pipeline),
99
        ("stable-diffusion-3-pag", StableDiffusion3PAGPipeline),
YiYi Xu's avatar
YiYi Xu committed
100
        ("if", IFPipeline),
101
        ("hunyuan", HunyuanDiTPipeline),
102
        ("hunyuan-pag", HunyuanDiTPAGPipeline),
103
104
        ("kandinsky", KandinskyCombinedPipeline),
        ("kandinsky22", KandinskyV22CombinedPipeline),
105
        ("kandinsky3", Kandinsky3Pipeline),
YiYi Xu's avatar
YiYi Xu committed
106
107
        ("stable-diffusion-controlnet", StableDiffusionControlNetPipeline),
        ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetPipeline),
Kashif Rasul's avatar
Kashif Rasul committed
108
        ("wuerstchen", WuerstchenCombinedPipeline),
109
        ("cascade", StableCascadeCombinedPipeline),
110
        ("lcm", LatentConsistencyModelPipeline),
111
112
        ("pixart-alpha", PixArtAlphaPipeline),
        ("pixart-sigma", PixArtSigmaPipeline),
113
        ("stable-diffusion-pag", StableDiffusionPAGPipeline),
114
        ("stable-diffusion-controlnet-pag", StableDiffusionControlNetPAGPipeline),
YiYi Xu's avatar
YiYi Xu committed
115
116
        ("stable-diffusion-xl-pag", StableDiffusionXLPAGPipeline),
        ("stable-diffusion-xl-controlnet-pag", StableDiffusionXLControlNetPAGPipeline),
117
        ("pixart-sigma-pag", PixArtSigmaPAGPipeline),
118
        ("auraflow", AuraFlowPipeline),
Sayak Paul's avatar
Sayak Paul committed
119
        ("flux", FluxPipeline),
120
        ("flux-controlnet", FluxControlNetPipeline),
121
        ("lumina", LuminaText2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
122
123
124
125
126
127
128
    ]
)

AUTO_IMAGE2IMAGE_PIPELINES_MAPPING = OrderedDict(
    [
        ("stable-diffusion", StableDiffusionImg2ImgPipeline),
        ("stable-diffusion-xl", StableDiffusionXLImg2ImgPipeline),
129
        ("stable-diffusion-3", StableDiffusion3Img2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
130
        ("if", IFImg2ImgPipeline),
131
132
        ("kandinsky", KandinskyImg2ImgCombinedPipeline),
        ("kandinsky22", KandinskyV22Img2ImgCombinedPipeline),
133
        ("kandinsky3", Kandinsky3Img2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
134
        ("stable-diffusion-controlnet", StableDiffusionControlNetImg2ImgPipeline),
135
        ("stable-diffusion-pag", StableDiffusionPAGImg2ImgPipeline),
136
        ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetImg2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
137
        ("stable-diffusion-xl-pag", StableDiffusionXLPAGImg2ImgPipeline),
138
        ("stable-diffusion-xl-controlnet-pag", StableDiffusionXLControlNetPAGImg2ImgPipeline),
139
        ("lcm", LatentConsistencyModelImg2ImgPipeline),
140
        ("flux", FluxImg2ImgPipeline),
141
        ("flux-controlnet", FluxControlNetImg2ImgPipeline),
YiYi Xu's avatar
YiYi Xu committed
142
143
144
145
146
147
148
    ]
)

AUTO_INPAINT_PIPELINES_MAPPING = OrderedDict(
    [
        ("stable-diffusion", StableDiffusionInpaintPipeline),
        ("stable-diffusion-xl", StableDiffusionXLInpaintPipeline),
149
        ("stable-diffusion-3", StableDiffusion3InpaintPipeline),
YiYi Xu's avatar
YiYi Xu committed
150
        ("if", IFInpaintingPipeline),
151
152
153
        ("kandinsky", KandinskyInpaintCombinedPipeline),
        ("kandinsky22", KandinskyV22InpaintCombinedPipeline),
        ("stable-diffusion-controlnet", StableDiffusionControlNetInpaintPipeline),
154
        ("stable-diffusion-controlnet-pag", StableDiffusionControlNetPAGInpaintPipeline),
155
        ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetInpaintPipeline),
YiYi Xu's avatar
YiYi Xu committed
156
        ("stable-diffusion-xl-pag", StableDiffusionXLPAGInpaintPipeline),
157
        ("flux", FluxInpaintPipeline),
158
        ("flux-controlnet", FluxControlNetInpaintPipeline),
159
160
161
162
163
164
165
    ]
)

_AUTO_TEXT2IMAGE_DECODER_PIPELINES_MAPPING = OrderedDict(
    [
        ("kandinsky", KandinskyPipeline),
        ("kandinsky22", KandinskyV22Pipeline),
Kashif Rasul's avatar
Kashif Rasul committed
166
        ("wuerstchen", WuerstchenDecoderPipeline),
167
        ("cascade", StableCascadeDecoderPipeline),
168
169
170
171
172
173
174
175
176
177
    ]
)
_AUTO_IMAGE2IMAGE_DECODER_PIPELINES_MAPPING = OrderedDict(
    [
        ("kandinsky", KandinskyImg2ImgPipeline),
        ("kandinsky22", KandinskyV22Img2ImgPipeline),
    ]
)
_AUTO_INPAINT_DECODER_PIPELINES_MAPPING = OrderedDict(
    [
YiYi Xu's avatar
YiYi Xu committed
178
179
180
181
182
        ("kandinsky", KandinskyInpaintPipeline),
        ("kandinsky22", KandinskyV22InpaintPipeline),
    ]
)

183
if is_sentencepiece_available():
184
    from .kolors import KolorsImg2ImgPipeline, KolorsPipeline
Álvaro Somoza's avatar
Álvaro Somoza committed
185
    from .pag import KolorsPAGPipeline
186
187

    AUTO_TEXT2IMAGE_PIPELINES_MAPPING["kolors"] = KolorsPipeline
Álvaro Somoza's avatar
Álvaro Somoza committed
188
    AUTO_TEXT2IMAGE_PIPELINES_MAPPING["kolors-pag"] = KolorsPAGPipeline
189
    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["kolors"] = KolorsImg2ImgPipeline
190

YiYi Xu's avatar
YiYi Xu committed
191
192
193
194
SUPPORTED_TASKS_MAPPINGS = [
    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
    AUTO_INPAINT_PIPELINES_MAPPING,
195
196
197
    _AUTO_TEXT2IMAGE_DECODER_PIPELINES_MAPPING,
    _AUTO_IMAGE2IMAGE_DECODER_PIPELINES_MAPPING,
    _AUTO_INPAINT_DECODER_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
198
199
200
]


201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
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
216
217
218
219
220
221
222
223
224
225
226
227
    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
228
229
230

    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
231
232
233
234
235


class AutoPipelineForText2Image(ConfigMixin):
    r"""

236
237
238
    [`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
239

240
    This class cannot be instantiated using `__init__()` (throws an error).
YiYi Xu's avatar
YiYi Xu committed
241
242
243
244
245
246
247

    Class attributes:

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

    """
248

YiYi Xu's avatar
YiYi Xu committed
249
250
251
252
253
254
255
256
257
258
    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
259
    @validate_hf_hub_args
YiYi Xu's avatar
YiYi Xu committed
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
    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:
283
            pretrained_model_or_path (`str` or `os.PathLike`, *optional*):
YiYi Xu's avatar
YiYi Xu committed
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
                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.
300

YiYi Xu's avatar
YiYi Xu committed
301
302
303
304
305
306
307
308
            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.
309
            token (`str` or *bool*, *optional*):
YiYi Xu's avatar
YiYi Xu committed
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
                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
367
        >>> from diffusers import AutoPipelineForText2Image
YiYi Xu's avatar
YiYi Xu committed
368

369
        >>> pipeline = AutoPipelineForText2Image.from_pretrained("runwayml/stable-diffusion-v1-5")
370
        >>> image = pipeline(prompt).images[0]
YiYi Xu's avatar
YiYi Xu committed
371
372
        ```
        """
373
        cache_dir = kwargs.pop("cache_dir", None)
374
375
        force_download = kwargs.pop("force_download", False)
        proxies = kwargs.pop("proxies", None)
376
        token = kwargs.pop("token", None)
377
378
379
380
381
382
383
        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,
384
            "token": token,
385
386
387
388
389
            "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
390
391
392
393
        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
394
395
396
397
        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
398
399
400

        text_2_image_cls = _get_task_class(AUTO_TEXT2IMAGE_PIPELINES_MAPPING, orig_class_name)

401
        kwargs = {**load_config_kwargs, **kwargs}
YiYi Xu's avatar
YiYi Xu committed
402
403
404
405
406
407
408
409
410
411
412
        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
413
        additional memory.
YiYi Xu's avatar
YiYi Xu committed
414
415
416
417
418
419
420
421

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

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

        ```py
422
        >>> from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
YiYi Xu's avatar
YiYi Xu committed
423
424
425
426
427

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

428
429
        >>> pipe_t2i = AutoPipelineForText2Image.from_pipe(pipe_i2i)
        >>> image = pipe_t2i(prompt).images[0]
YiYi Xu's avatar
YiYi Xu committed
430
431
432
433
434
435
436
437
438
        ```
        """

        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)

439
440
        if "controlnet" in kwargs:
            if kwargs["controlnet"] is not None:
YiYi Xu's avatar
YiYi Xu committed
441
                to_replace = "PAGPipeline" if "PAG" in text_2_image_cls.__name__ else "Pipeline"
442
443
                text_2_image_cls = _get_task_class(
                    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
444
                    text_2_image_cls.__name__.replace("ControlNet", "").replace(to_replace, "ControlNet" + to_replace),
445
446
447
448
                )
            else:
                text_2_image_cls = _get_task_class(
                    AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
449
450
451
452
453
454
455
456
457
458
459
460
461
462
                    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", ""),
463
464
                )

YiYi Xu's avatar
YiYi Xu committed
465
        # define expected module and optional kwargs given the pipeline signature
466
        expected_modules, optional_kwargs = text_2_image_cls._get_signature_keys(text_2_image_cls)
YiYi Xu's avatar
YiYi Xu committed
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508

        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(
509
                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
510
511
512
513
514
515
516
517
518
519
520
521
            )

        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"""

522
523
524
    [`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
525

526
    This class cannot be instantiated using `__init__()` (throws an error).
YiYi Xu's avatar
YiYi Xu committed
527
528
529
530
531
532
533

    Class attributes:

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

    """
534

YiYi Xu's avatar
YiYi Xu committed
535
536
537
538
539
540
541
542
543
544
    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
545
    @validate_hf_hub_args
YiYi Xu's avatar
YiYi Xu committed
546
547
548
549
550
551
552
553
554
555
    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.

556
557
        If a `controlnet` argument is passed, it will instantiate a [`StableDiffusionControlNetImg2ImgPipeline`]
        object.
YiYi Xu's avatar
YiYi Xu committed
558
559
560
561
562
563
564
565
566
567
568
569

        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:
570
            pretrained_model_or_path (`str` or `os.PathLike`, *optional*):
YiYi Xu's avatar
YiYi Xu committed
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
                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.
587

YiYi Xu's avatar
YiYi Xu committed
588
589
590
591
592
593
594
595
            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.
596
            token (`str` or *bool*, *optional*):
YiYi Xu's avatar
YiYi Xu committed
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
                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
654
        >>> from diffusers import AutoPipelineForImage2Image
YiYi Xu's avatar
YiYi Xu committed
655

656
657
        >>> pipeline = AutoPipelineForImage2Image.from_pretrained("runwayml/stable-diffusion-v1-5")
        >>> image = pipeline(prompt, image).images[0]
YiYi Xu's avatar
YiYi Xu committed
658
659
        ```
        """
660
        cache_dir = kwargs.pop("cache_dir", None)
661
662
        force_download = kwargs.pop("force_download", False)
        proxies = kwargs.pop("proxies", None)
663
        token = kwargs.pop("token", None)
664
665
666
667
668
669
670
        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,
671
            "token": token,
672
673
674
675
676
            "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
677
678
        orig_class_name = config["_class_name"]

679
680
681
682
683
        # the `orig_class_name` can be:
        # `- *Pipeline` (for regular text-to-image checkpoint)
        # `- *Img2ImgPipeline` (for refiner checkpoint)
        to_replace = "Img2ImgPipeline" if "Img2Img" in config["_class_name"] else "Pipeline"

YiYi Xu's avatar
YiYi Xu committed
684
        if "controlnet" in kwargs:
685
            orig_class_name = orig_class_name.replace(to_replace, "ControlNet" + to_replace)
YiYi Xu's avatar
YiYi Xu committed
686
687
688
        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
689
                orig_class_name = orig_class_name.replace(to_replace, "PAG" + to_replace)
YiYi Xu's avatar
YiYi Xu committed
690
691
692

        image_2_image_cls = _get_task_class(AUTO_IMAGE2IMAGE_PIPELINES_MAPPING, orig_class_name)

693
        kwargs = {**load_config_kwargs, **kwargs}
YiYi Xu's avatar
YiYi Xu committed
694
695
696
697
698
699
700
701
702
703
704
        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
705
        additional memory.
YiYi Xu's avatar
YiYi Xu committed
706
707
708
709
710
711
712
713
714
715

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

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

        Examples:

        ```py
716
        >>> from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
YiYi Xu's avatar
YiYi Xu committed
717
718
719
720
721

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

722
723
        >>> pipe_i2i = AutoPipelineForImage2Image.from_pipe(pipe_t2i)
        >>> image = pipe_i2i(prompt, image).images[0]
YiYi Xu's avatar
YiYi Xu committed
724
725
726
727
728
729
730
731
732
        ```
        """

        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)

733
734
        if "controlnet" in kwargs:
            if kwargs["controlnet"] is not None:
YiYi Xu's avatar
YiYi Xu committed
735
736
737
                to_replace = "Img2ImgPipeline"
                if "PAG" in image_2_image_cls.__name__:
                    to_replace = "PAG" + to_replace
738
739
                image_2_image_cls = _get_task_class(
                    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
740
                    image_2_image_cls.__name__.replace("ControlNet", "").replace(
YiYi Xu's avatar
YiYi Xu committed
741
                        to_replace, "ControlNet" + to_replace
742
                    ),
743
744
745
746
                )
            else:
                image_2_image_cls = _get_task_class(
                    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING,
YiYi Xu's avatar
YiYi Xu committed
747
748
749
750
751
752
753
754
755
756
757
758
759
760
                    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", ""),
761
762
                )

YiYi Xu's avatar
YiYi Xu committed
763
        # define expected module and optional kwargs given the pipeline signature
764
        expected_modules, optional_kwargs = image_2_image_cls._get_signature_keys(image_2_image_cls)
YiYi Xu's avatar
YiYi Xu committed
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806

        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(
807
                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
808
809
810
811
812
813
814
815
816
817
818
819
            )

        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"""

820
821
822
    [`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
823

824
    This class cannot be instantiated using `__init__()` (throws an error).
YiYi Xu's avatar
YiYi Xu committed
825
826
827
828
829
830
831

    Class attributes:

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

    """
832

YiYi Xu's avatar
YiYi Xu committed
833
834
835
836
837
838
839
840
841
842
    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
843
    @validate_hf_hub_args
YiYi Xu's avatar
YiYi Xu committed
844
845
846
847
848
849
850
851
852
    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.

853
854
        If a `controlnet` argument is passed, it will instantiate a [`StableDiffusionControlNetInpaintPipeline`]
        object.
YiYi Xu's avatar
YiYi Xu committed
855
856
857
858
859
860
861
862
863
864
865
866

        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:
867
            pretrained_model_or_path (`str` or `os.PathLike`, *optional*):
YiYi Xu's avatar
YiYi Xu committed
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
                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.
884

YiYi Xu's avatar
YiYi Xu committed
885
886
887
888
889
890
891
892
            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.
893
            token (`str` or *bool*, *optional*):
YiYi Xu's avatar
YiYi Xu committed
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
                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
951
        >>> from diffusers import AutoPipelineForInpainting
YiYi Xu's avatar
YiYi Xu committed
952

953
954
        >>> 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
955
956
        ```
        """
957
        cache_dir = kwargs.pop("cache_dir", None)
958
959
        force_download = kwargs.pop("force_download", False)
        proxies = kwargs.pop("proxies", None)
960
        token = kwargs.pop("token", None)
961
962
963
964
965
966
967
        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,
968
            "token": token,
969
970
971
972
973
            "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
974
975
        orig_class_name = config["_class_name"]

976
977
978
979
980
        # The `orig_class_name`` can be:
        # `- *InpaintPipeline` (for inpaint-specific checkpoint)
        #  - or *Pipeline (for regular text-to-image checkpoint)
        to_replace = "InpaintPipeline" if "Inpaint" in config["_class_name"] else "Pipeline"

YiYi Xu's avatar
YiYi Xu committed
981
        if "controlnet" in kwargs:
982
            orig_class_name = orig_class_name.replace(to_replace, "ControlNet" + to_replace)
YiYi Xu's avatar
YiYi Xu committed
983
984
985
        if "enable_pag" in kwargs:
            enable_pag = kwargs.pop("enable_pag")
            if enable_pag:
986
                orig_class_name = orig_class_name.replace(to_replace, "PAG" + to_replace)
YiYi Xu's avatar
YiYi Xu committed
987
988
        inpainting_cls = _get_task_class(AUTO_INPAINT_PIPELINES_MAPPING, orig_class_name)

989
        kwargs = {**load_config_kwargs, **kwargs}
YiYi Xu's avatar
YiYi Xu committed
990
991
992
993
994
995
996
997
998
999
1000
        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
1001
        additional memory.
YiYi Xu's avatar
YiYi Xu committed
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011

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

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

        Examples:

        ```py
1012
        >>> from diffusers import AutoPipelineForText2Image, AutoPipelineForInpainting
YiYi Xu's avatar
YiYi Xu committed
1013
1014
1015
1016
1017
1018

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

        >>> pipe_inpaint = AutoPipelineForInpainting.from_pipe(pipe_t2i)
1019
        >>> image = pipe_inpaint(prompt, image=init_image, mask_image=mask_image).images[0]
YiYi Xu's avatar
YiYi Xu committed
1020
1021
1022
1023
1024
1025
1026
1027
        ```
        """
        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)

1028
1029
1030
1031
        if "controlnet" in kwargs:
            if kwargs["controlnet"] is not None:
                inpainting_cls = _get_task_class(
                    AUTO_INPAINT_PIPELINES_MAPPING,
1032
1033
1034
                    inpainting_cls.__name__.replace("ControlNet", "").replace(
                        "InpaintPipeline", "ControlNetInpaintPipeline"
                    ),
1035
1036
1037
1038
1039
1040
1041
                )
            else:
                inpainting_cls = _get_task_class(
                    AUTO_INPAINT_PIPELINES_MAPPING,
                    inpainting_cls.__name__.replace("ControlNetInpaintPipeline", "InpaintPipeline"),
                )

YiYi Xu's avatar
YiYi Xu committed
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
        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
1055
        # define expected module and optional kwargs given the pipeline signature
1056
        expected_modules, optional_kwargs = inpainting_cls._get_signature_keys(inpainting_cls)
YiYi Xu's avatar
YiYi Xu committed
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098

        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(
1099
                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
1100
1101
1102
1103
1104
1105
1106
            )

        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