add autopipeline for text2video task

Author: naomili0924

src/diffusers/__init__.py

@@ -306,6 +306,7 @@
             "AutoPipelineForImage2Image",
             "AutoPipelineForInpainting",
             "AutoPipelineForText2Image",
+            "AutoPipelineForText2Video",
             "ConsistencyModelPipeline",
             "DanceDiffusionPipeline",
             "DDIMPipeline",

src/diffusers/pipelines/__init__.py

@@ -46,6 +46,7 @@
         "AutoPipelineForImage2Image",
         "AutoPipelineForInpainting",
         "AutoPipelineForText2Image",
+        "AutoPipelineForText2Video",
     ]
     _import_structure["consistency_models"] = ["ConsistencyModelPipeline"]
     _import_structure["dance_diffusion"] = ["DanceDiffusionPipeline"]

src/diffusers/pipelines/auto_pipeline.py

@@ -117,7 +117,8 @@
     StableDiffusionXLInpaintPipeline,
     StableDiffusionXLPipeline,
 )
-from .wan import WanImageToVideoPipeline, WanPipeline, WanVideoToVideoPipeline
+from .text_to_video_synthesis import TextToVideoSDPipeline
+from .wan import WanAnimatePipeline, WanImageToVideoPipeline, WanPipeline, WanVACEPipeline, WanVideoToVideoPipeline
 from .wuerstchen import WuerstchenCombinedPipeline, WuerstchenDecoderPipeline
 from .z_image import ZImageImg2ImgPipeline, ZImagePipeline
 
@@ -221,6 +222,9 @@
 AUTO_TEXT2VIDEO_PIPELINES_MAPPING = OrderedDict(
     [
         ("wan", WanPipeline),
+        ("wan-animate", WanAnimatePipeline),
+        ("wan-vace", WanVACEPipeline),
+        ("stable-diffusion", TextToVideoSDPipeline),
     ]
 )
 
@@ -1206,3 +1210,246 @@ def from_pipe(cls, pipeline, **kwargs):
         model.register_to_config(**unused_original_config)
 
         return model
+
+
+class AutoPipelineForText2Video(ConfigMixin):
+    r"""
+
+    [`AutoPipelineForText2Video`] is a generic pipeline class that instantiates an text-to-video pipeline class. The
+    specific underlying pipeline class is automatically selected from either the
+    [`~AutoPipelineForText2Video.from_pretrained`] or [`~AutoPipelineForText2Video.from_pipe`] methods.
+
+    This class cannot be instantiated using `__init__()` (throws an error).
+
+    Class attributes:
+
+        - **config_name** (`str`) -- The configuration filename that stores the class and module names of all the
+          diffusion pipeline's components.
+
+    """
+
+    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
+    @validate_hf_hub_args
+    def from_pretrained(cls, pretrained_model_or_path, **kwargs):
+        r"""
+        Instantiates a text-to-video 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-video pipeline linked to the pipeline class using pattern matching on pipeline class name.
+
+
+        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 stable-diffusion-v1-5/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:
+            pretrained_model_or_path (`str` or `os.PathLike`, *optional*):
+                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.
+            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.
+
+            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.
+            token (`str` or *bool*, *optional*):
+                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 `hf > auth login`.
+
+        Examples:
+
+        ```py
+        >>> from diffusers import AutoPipelineForText2Video
+
+        >>> pipeline = AutoPipelineForText2Video.from_pretrained("damo-vilab/text-to-video-ms-1.7b")
+        >>> video_frames = pipe(prompt, num_frames=32).frames[0]
+        ```
+        """
+
+        cache_dir = kwargs.pop("cache_dir", None)
+        force_download = kwargs.pop("force_download", False)
+        proxies = kwargs.pop("proxies", None)
+        token = kwargs.pop("token", None)
+        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,
+            "token": token,
+            "local_files_only": local_files_only,
+            "revision": revision,
+        }
+
+        config = cls.load_config(pretrained_model_or_path, **load_config_kwargs)
+        orig_class_name = config["_class_name"]
+        text_to_video_cls = _get_task_class(AUTO_TEXT2VIDEO_PIPELINES_MAPPING, orig_class_name)
+        kwargs = {**load_config_kwargs, **kwargs}
+        return text_to_video_cls.from_pretrained(pretrained_model_or_path, **kwargs)
+
+    @classmethod
+    def from_pipe(cls, pipeline, **kwargs):
+        r"""
+        Instantiates a text-to-video 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-video
+        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
+        additional memory.
+
+        The pipeline is set in evaluation mode (`model.eval()`) by default.
+
+        Parameters:
+            pipeline (`DiffusionPipeline`):
+                an instantiated `DiffusionPipeline` object
+
+        Examples:
+
+        ```py
+        >>> from diffusers import AutoPipelineForText2Video
+        >>> pipeline = AutoPipelineForText2Video.from_pretrained("damo-vilab/text-to-video-ms-1.7b")
+        >>> output = pipeline(prompt, negative_prompt=negative_prompt, height=height, width=width, num_frames=num_frames).frames[0]
+        ```
+        """
+        original_config = dict(pipeline.config)
+        original_cls_name = pipeline.__class__.__name__
+
+        # derive the pipeline class to instantiate
+        text_to_video_cls = _get_task_class(AUTO_TEXT2VIDEO_PIPELINES_MAPPING, original_cls_name)
+
+        # define expected module and optional kwargs given the pipeline signature
+        expected_modules, optional_kwargs = text_to_video_cls._get_signature_keys(text_to_video_cls)
+
+        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_to_video_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_to_video_kwargs
+        }
+
+        missing_modules = (
+            set(expected_modules) - set(text_to_video_cls._optional_components) - set(text_to_video_kwargs.keys())
+        )
+
+        if len(missing_modules) > 0:
+            raise ValueError(
+                f"Pipeline {text_to_video_cls} expected {expected_modules}, but only {set(list(passed_class_obj.keys()) + list(original_class_obj.keys()))} were passed"
+            )
+
+        model = text_to_video_cls(**text_to_video_kwargs)
+        model.register_to_config(_name_or_path=pretrained_model_name_or_path)
+        model.register_to_config(**unused_original_config)
+
+        return model

tests/pipelines/test_pipelines_auto.py

@@ -27,6 +27,7 @@
     AutoPipelineForImage2Image,
     AutoPipelineForInpainting,
     AutoPipelineForText2Image,
+    AutoPipelineForText2Video,
     ControlNetModel,
     DiffusionPipeline,
 )
@@ -455,6 +456,15 @@ def test_from_pipe_optional_components(self):
         pipe = AutoPipelineForText2Image.from_pipe(pipe, image_encoder=None)
         assert pipe.image_encoder is None
 
+    def test_from_pretrained_text_to_video(self):
+        repo = "hf-internal-testing/tiny-stable-diffusion-pipe"
+
+        pipe = AutoPipelineForText2Video.from_pretrained(repo)
+        assert pipe.__class__.__name__ == "TextToVideoSDPipeline"
+
+        pipe = AutoPipelineForText2Video.from_pipe(pipe)
+        assert pipe.__class__.__name__ == "TextToVideoSDPipeline"
+
 
 @slow
 class AutoPipelineIntegrationTest(unittest.TestCase):