Stable Video Diffusion

Stable Video Diffusion was proposed in Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets by Andreas Blattmann, Tim Dockhorn, Sumith Kulal, Daniel Mendelevitch, Maciej Kilian, Dominik Lorenz, Yam Levi, Zion English, Vikram Voleti, Adam Letts, Varun Jampani, Robin Rombach.

The abstract from the paper is:

We present Stable Video Diffusion - a latent video diffusion model for high-resolution, state-of-the-art text-to-video and image-to-video generation. Recently, latent diffusion models trained for 2D image synthesis have been turned into generative video models by inserting temporal layers and finetuning them on small, high-quality video datasets. However, training methods in the literature vary widely, and the field has yet to agree on a unified strategy for curating video data. In this paper, we identify and evaluate three different stages for successful training of video LDMs: text-to-image pretraining, video pretraining, and high-quality video finetuning. Furthermore, we demonstrate the necessity of a well-curated pretraining dataset for generating high-quality videos and present a systematic curation process to train a strong base model, including captioning and filtering strategies. We then explore the impact of finetuning our base model on high-quality data and train a text-to-video model that is competitive with closed-source video generation. We also show that our base model provides a powerful motion representation for downstream tasks such as image-to-video generation and adaptability to camera motion-specific LoRA modules. Finally, we demonstrate that our model provides a strong multi-view 3D-prior and can serve as a base to finetune a multi-view diffusion model that jointly generates multiple views of objects in a feedforward fashion, outperforming image-based methods at a fraction of their compute budget. We release code and model weights at this https URL.

Tip

To learn how to use Stable Video Diffusion, take a look at the Stable Video Diffusion guide.

Check out the Stability AI Hub organization for the base and extended frame checkpoints!

Tips

Video generation is memory-intensive and one way to reduce your memory usage is to set enable_forward_chunking on the pipeline's UNet so you don't run the entire feedforward layer at once. Breaking it up into chunks in a loop is more efficient.

Check out the Text or image-to-video guide for more details about how certain parameters can affect video generation and how to optimize inference by reducing memory usage.

mindone.diffusers.StableVideoDiffusionPipeline

Bases: DiffusionPipeline

Pipeline to generate video from an input image using Stable Video Diffusion.

This model inherits from [DiffusionPipeline]. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.).

PARAMETER	DESCRIPTION
vae	Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations. TYPE: [`AutoencoderKLTemporalDecoder`]
image_encoder	Frozen CLIP image-encoder (laion/CLIP-ViT-H-14-laion2B-s32B-b79K). TYPE: [`~transformers.CLIPVisionModelWithProjection`]
unet	A UNetSpatioTemporalConditionModel to denoise the encoded image latents. TYPE: [`UNetSpatioTemporalConditionModel`]
scheduler	A scheduler to be used in combination with unet to denoise the encoded image latents. TYPE: [`EulerDiscreteScheduler`]
feature_extractor	A CLIPImageProcessor to extract features from generated images. TYPE: [`~transformers.CLIPImageProcessor`]

Source code in mindone/diffusers/pipelines/stable_video_diffusion/pipeline_stable_video_diffusion.py

class StableVideoDiffusionPipeline(DiffusionPipeline):
    r"""
    Pipeline to generate video from an input image using Stable Video Diffusion.

    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
    implemented for all pipelines (downloading, saving, running on a particular device, etc.).

    Args:
        vae ([`AutoencoderKLTemporalDecoder`]):
            Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
        image_encoder ([`~transformers.CLIPVisionModelWithProjection`]):
            Frozen CLIP image-encoder
            ([laion/CLIP-ViT-H-14-laion2B-s32B-b79K](https://huggingface.co/laion/CLIP-ViT-H-14-laion2B-s32B-b79K)).
        unet ([`UNetSpatioTemporalConditionModel`]):
            A `UNetSpatioTemporalConditionModel` to denoise the encoded image latents.
        scheduler ([`EulerDiscreteScheduler`]):
            A scheduler to be used in combination with `unet` to denoise the encoded image latents.
        feature_extractor ([`~transformers.CLIPImageProcessor`]):
            A `CLIPImageProcessor` to extract features from generated images.
    """

    model_cpu_offload_seq = "image_encoder->unet->vae"
    _callback_tensor_inputs = ["latents"]

    def __init__(
        self,
        vae: AutoencoderKLTemporalDecoder,
        image_encoder: CLIPVisionModelWithProjection,
        unet: UNetSpatioTemporalConditionModel,
        scheduler: EulerDiscreteScheduler,
        feature_extractor: CLIPImageProcessor,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            image_encoder=image_encoder,
            unet=unet,
            scheduler=scheduler,
            feature_extractor=feature_extractor,
        )
        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
        self.video_processor = VideoProcessor(do_resize=True, vae_scale_factor=self.vae_scale_factor)

    def _encode_image(
        self,
        image: PipelineImageInput,
        num_videos_per_prompt: int,
        do_classifier_free_guidance: bool,
    ) -> ms.Tensor:
        dtype = next(self.image_encoder.get_parameters()).dtype

        if not isinstance(image, ms.Tensor):
            image = self.video_processor.pil_to_numpy(image)
            image = self.video_processor.numpy_to_ms(image)

            # We normalize the image before resizing to match with the original implementation.
            # Then we unnormalize it after resizing.
            image = image * 2.0 - 1.0
            image = _resize_with_antialiasing(image, (224, 224))
            image = (image + 1.0) / 2.0

        # Normalize the image with for CLIP input
        image = self.feature_extractor(
            images=image.asnumpy(),
            do_normalize=True,
            do_center_crop=False,
            do_resize=False,
            do_rescale=False,
            return_tensors="np",
        ).pixel_values

        image = ms.tensor(image).to(dtype=dtype)
        image_embeddings = self.image_encoder(image)[0]
        image_embeddings = image_embeddings.unsqueeze(1)

        # duplicate image embeddings for each generation per prompt, using mps friendly method
        bs_embed, seq_len, _ = image_embeddings.shape
        image_embeddings = image_embeddings.tile((1, num_videos_per_prompt, 1))
        image_embeddings = image_embeddings.view(bs_embed * num_videos_per_prompt, seq_len, -1)

        if do_classifier_free_guidance:
            negative_image_embeddings = ops.zeros_like(image_embeddings)

            # For classifier free guidance, we need to do two forward passes.
            # Here we concatenate the unconditional and text embeddings into a single batch
            # to avoid doing two forward passes
            image_embeddings = ops.cat([negative_image_embeddings, image_embeddings])

        return image_embeddings

    def _encode_vae_image(
        self,
        image: ms.Tensor,
        num_videos_per_prompt: int,
        do_classifier_free_guidance: bool,
    ):
        image_latents = self.vae.diag_gauss_dist.mode(self.vae.encode(image)[0])

        # duplicate image_latents for each generation per prompt, using mps friendly method
        image_latents = image_latents.tile((num_videos_per_prompt, 1, 1, 1))

        if do_classifier_free_guidance:
            negative_image_latents = ops.zeros_like(image_latents)

            # For classifier free guidance, we need to do two forward passes.
            # Here we concatenate the unconditional and text embeddings into a single batch
            # to avoid doing two forward passes
            image_latents = ops.cat([negative_image_latents, image_latents])

        return image_latents

    def _get_add_time_ids(
        self,
        fps: int,
        motion_bucket_id: int,
        noise_aug_strength: float,
        dtype: ms.dtype,
        batch_size: int,
        num_videos_per_prompt: int,
        do_classifier_free_guidance: bool,
    ):
        add_time_ids = [fps, motion_bucket_id, noise_aug_strength]

        passed_add_embed_dim = self.unet.config.addition_time_embed_dim * len(add_time_ids)
        expected_add_embed_dim = self.unet.add_embedding.linear_1.in_channels

        if expected_add_embed_dim != passed_add_embed_dim:
            raise ValueError(
                f"Model expects an added time embedding vector of length {expected_add_embed_dim}, "
                f"but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. "
                "Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
            )

        add_time_ids = ms.Tensor([add_time_ids], dtype=dtype)
        add_time_ids = add_time_ids.tile((batch_size * num_videos_per_prompt, 1))

        if do_classifier_free_guidance:
            add_time_ids = ops.cat([add_time_ids, add_time_ids])

        return add_time_ids

    def decode_latents(self, latents: ms.Tensor, num_frames: int, decode_chunk_size: int = 14):
        # [batch, frames, channels, height, width] -> [batch*frames, channels, height, width]
        latents = latents.flatten(start_dim=0, end_dim=1)

        latents = 1 / self.vae.config.scaling_factor * latents

        forward_vae_fn = self.vae.construct
        accepts_num_frames = "num_frames" in set(inspect.signature(forward_vae_fn).parameters.keys())

        # decode decode_chunk_size frames at a time to avoid OOM
        frames = []
        for i in range(0, latents.shape[0], decode_chunk_size):
            num_frames_in = latents[i : i + decode_chunk_size].shape[0]
            decode_kwargs = {}
            if accepts_num_frames:
                # we only pass num_frames_in if it's expected
                decode_kwargs["num_frames"] = num_frames_in

            frame = self.vae.decode(latents[i : i + decode_chunk_size], **decode_kwargs)[0]
            frames.append(frame)
        frames = ops.cat(frames, axis=0)

        # [batch*frames, channels, height, width] -> [batch, channels, frames, height, width]
        frames = frames.reshape(-1, num_frames, *frames.shape[1:]).permute((0, 2, 1, 3, 4))

        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
        frames = frames.float()
        return frames

    def check_inputs(self, image, height, width):
        if not isinstance(image, ms.Tensor) and not isinstance(image, PIL.Image.Image) and not isinstance(image, list):
            raise ValueError(
                "`image` has to be of type `ms.Tensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is"
                f" {type(image)}"
            )

        if height % 8 != 0 or width % 8 != 0:
            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")

    def prepare_latents(
        self,
        batch_size: int,
        num_frames: int,
        num_channels_latents: int,
        height: int,
        width: int,
        dtype: ms.dtype,
        generator: np.random.Generator,
        latents: Optional[ms.Tensor] = None,
    ):
        shape = (
            batch_size,
            num_frames,
            num_channels_latents // 2,
            height // self.vae_scale_factor,
            width // self.vae_scale_factor,
        )
        if isinstance(generator, list) and len(generator) != batch_size:
            raise ValueError(
                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
            )

        if latents is None:
            latents = randn_tensor(shape, generator=generator, dtype=dtype)

        # scale the initial noise by the standard deviation required by the scheduler
        latents = (latents * self.scheduler.init_noise_sigma).to(dtype)
        return latents

    @property
    def guidance_scale(self):
        return self._guidance_scale

    # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
    # corresponds to doing no classifier free guidance.
    @property
    def do_classifier_free_guidance(self):
        if isinstance(self.guidance_scale, (int, float)):
            return self.guidance_scale > 1
        return self.guidance_scale.max() > 1

    @property
    def num_timesteps(self):
        return self._num_timesteps

    def __call__(
        self,
        image: Union[PIL.Image.Image, List[PIL.Image.Image], ms.Tensor],
        height: int = 576,
        width: int = 1024,
        num_frames: Optional[int] = None,
        num_inference_steps: int = 25,
        sigmas: Optional[List[float]] = None,
        min_guidance_scale: float = 1.0,
        max_guidance_scale: float = 3.0,
        fps: int = 7,
        motion_bucket_id: int = 127,
        noise_aug_strength: float = 0.02,
        decode_chunk_size: Optional[int] = None,
        num_videos_per_prompt: Optional[int] = 1,
        generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
        latents: Optional[ms.Tensor] = None,
        output_type: Optional[str] = "pil",
        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
        return_dict: bool = False,
    ):
        r"""
        The call function to the pipeline for generation.

        Args:
            image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `ms.Tensor`):
                Image(s) to guide image generation. If you provide a tensor, the expected value range is between `[0, 1]`.
            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The height in pixels of the generated image.
            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
                The width in pixels of the generated image.
            num_frames (`int`, *optional*):
                The number of video frames to generate. Defaults to `self.unet.config.num_frames`
                (14 for `stable-video-diffusion-img2vid` and to 25 for `stable-video-diffusion-img2vid-xt`).
            num_inference_steps (`int`, *optional*, defaults to 25):
                The number of denoising steps. More denoising steps usually lead to a higher quality video at the
                expense of slower inference. This parameter is modulated by `strength`.
            sigmas (`List[float]`, *optional*):
                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                will be used.
            min_guidance_scale (`float`, *optional*, defaults to 1.0):
                The minimum guidance scale. Used for the classifier free guidance with first frame.
            max_guidance_scale (`float`, *optional*, defaults to 3.0):
                The maximum guidance scale. Used for the classifier free guidance with last frame.
            fps (`int`, *optional*, defaults to 7):
                Frames per second. The rate at which the generated images shall be exported to a video after generation.
                Note that Stable Diffusion Video's UNet was micro-conditioned on fps-1 during training.
            motion_bucket_id (`int`, *optional*, defaults to 127):
                Used for conditioning the amount of motion for the generation. The higher the number the more motion
                will be in the video.
            noise_aug_strength (`float`, *optional*, defaults to 0.02):
                The amount of noise added to the init image, the higher it is the less the video will look like the init image. Increase it for more motion.
            decode_chunk_size (`int`, *optional*):
                The number of frames to decode at a time. Higher chunk size leads to better temporal consistency at the expense of more memory usage.
                By default, the decoder decodes all frames at once for maximal quality. For lower memory usage, reduce `decode_chunk_size`.
            num_videos_per_prompt (`int`, *optional*, defaults to 1):
                The number of videos to generate per prompt.
            generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
                A [`np.random.Generator`](https://numpy.org/doc/stable/reference/random/generator.html) to make
                generation deterministic.
            latents (`ms.Tensor`, *optional*):
                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for video
                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
                tensor is generated by sampling using the supplied random `generator`.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generated image. Choose between `pil`, `np` or `ms`.
            callback_on_step_end (`Callable`, *optional*):
                A function that is called at the end of each denoising step during inference. The function is called
                with the following arguments:
                    `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`.
                `callback_kwargs` will include a list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
            callback_on_step_end_tensor_inputs (`List`, *optional*):
                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
                `._callback_tensor_inputs` attribute of your pipeline class.
            return_dict (`bool`, *optional*, defaults to `False`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.

        Examples:

        Returns:
            [`~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput`] or `tuple`:
                If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput`] is returned,
                otherwise a `tuple` of (`List[List[PIL.Image.Image]]` or `np.ndarray` or `ms.Tensor`) is returned.
        """
        # 0. Default height and width to unet
        height = height or self.unet.config.sample_size * self.vae_scale_factor
        width = width or self.unet.config.sample_size * self.vae_scale_factor

        num_frames = num_frames if num_frames is not None else self.unet.config.num_frames
        decode_chunk_size = decode_chunk_size if decode_chunk_size is not None else num_frames

        # 1. Check inputs. Raise error if not correct
        self.check_inputs(image, height, width)

        # 2. Define call parameters
        if isinstance(image, PIL.Image.Image):
            batch_size = 1
        elif isinstance(image, list):
            batch_size = len(image)
        else:
            batch_size = image.shape[0]
        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
        # corresponds to doing no classifier free guidance.
        self._guidance_scale = max_guidance_scale

        # 3. Encode input image
        image_embeddings = self._encode_image(image, num_videos_per_prompt, self.do_classifier_free_guidance)

        # NOTE: Stable Video Diffusion was conditioned on fps - 1, which is why it is reduced here.
        # See: https://github.com/Stability-AI/generative-models/blob/ed0997173f98eaf8f4edf7ba5fe8f15c6b877fd3/scripts/sampling/simple_video_sample.py#L188
        fps = fps - 1

        # 4. Encode input image using VAE
        image = self.video_processor.preprocess(image, height=height, width=width)
        noise = randn_tensor(image.shape, generator=generator, dtype=image.dtype)
        image = image + noise_aug_strength * noise

        needs_upcasting = self.vae.dtype == ms.float16 and self.vae.config.force_upcast
        if needs_upcasting:
            self.vae.to(dtype=ms.float32)

        image_latents = self._encode_vae_image(
            image,
            num_videos_per_prompt=num_videos_per_prompt,
            do_classifier_free_guidance=self.do_classifier_free_guidance,
        )
        image_latents = image_latents.to(image_embeddings.dtype)

        # cast back to fp16 if needed
        if needs_upcasting:
            self.vae.to(dtype=ms.float16)

        # Repeat the image latents for each frame so we can concatenate them with the noise
        # image_latents [batch, channels, height, width] ->[batch, num_frames, channels, height, width]
        image_latents = image_latents.unsqueeze(1).tile((1, num_frames, 1, 1, 1))

        # 5. Get Added Time IDs
        added_time_ids = self._get_add_time_ids(
            fps,
            motion_bucket_id,
            noise_aug_strength,
            image_embeddings.dtype,
            batch_size,
            num_videos_per_prompt,
            self.do_classifier_free_guidance,
        )

        # 6. Prepare timesteps
        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, None, sigmas)

        # 7. Prepare latent variables
        num_channels_latents = self.unet.config.in_channels
        latents = self.prepare_latents(
            batch_size * num_videos_per_prompt,
            num_frames,
            num_channels_latents,
            height,
            width,
            image_embeddings.dtype,
            generator,
            latents,
        )

        # 8. Prepare guidance scale
        guidance_scale = ms.Tensor.from_numpy(
            np.linspace(min_guidance_scale, max_guidance_scale, num_frames)
        ).unsqueeze(0)
        guidance_scale = guidance_scale.to(latents.dtype)
        guidance_scale = guidance_scale.tile((batch_size * num_videos_per_prompt, 1))
        guidance_scale = _append_dims(guidance_scale, latents.ndim)

        self._guidance_scale = guidance_scale

        # 9. Denoising loop
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        self._num_timesteps = len(timesteps)
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                # expand the latents if we are doing classifier free guidance
                latent_model_input = ops.cat([latents] * 2) if self.do_classifier_free_guidance else latents
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

                # Concatenate image_latents over channels dimension
                latent_model_input = ops.cat([latent_model_input, image_latents], axis=2)

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=image_embeddings,
                    added_time_ids=added_time_ids,
                    return_dict=False,
                )[0]

                # perform guidance
                if self.do_classifier_free_guidance:
                    noise_pred_uncond, noise_pred_cond = noise_pred.chunk(2)
                    noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_cond - noise_pred_uncond)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents)[0]

                if callback_on_step_end is not None:
                    callback_kwargs = {}
                    for k in callback_on_step_end_tensor_inputs:
                        callback_kwargs[k] = locals()[k]
                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                    latents = callback_outputs.pop("latents", latents)

                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        if not output_type == "latent":
            # cast back to fp16 if needed
            if needs_upcasting:
                self.vae.to(dtype=ms.float16)
            frames = self.decode_latents(latents, num_frames, decode_chunk_size)
            frames = self.video_processor.postprocess_video(video=frames, output_type=output_type)
        else:
            frames = latents

        if not return_dict:
            return frames

        return StableVideoDiffusionPipelineOutput(frames=frames)

mindone.diffusers.StableVideoDiffusionPipeline.__call__(image, height=576, width=1024, num_frames=None, num_inference_steps=25, sigmas=None, min_guidance_scale=1.0, max_guidance_scale=3.0, fps=7, motion_bucket_id=127, noise_aug_strength=0.02, decode_chunk_size=None, num_videos_per_prompt=1, generator=None, latents=None, output_type='pil', callback_on_step_end=None, callback_on_step_end_tensor_inputs=['latents'], return_dict=False)

The call function to the pipeline for generation.

PARAMETER	DESCRIPTION
image	Image(s) to guide image generation. If you provide a tensor, the expected value range is between [0, 1]. TYPE: `PIL.Image.Image` or `List[PIL.Image.Image]` or `ms.Tensor`
height	The height in pixels of the generated image. TYPE: `int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor` DEFAULT: 576
width	The width in pixels of the generated image. TYPE: `int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor` DEFAULT: 1024
num_frames	The number of video frames to generate. Defaults to self.unet.config.num_frames (14 for stable-video-diffusion-img2vid and to 25 for stable-video-diffusion-img2vid-xt). TYPE: `int`, *optional* DEFAULT: None
num_inference_steps	The number of denoising steps. More denoising steps usually lead to a higher quality video at the expense of slower inference. This parameter is modulated by strength. TYPE: `int`, *optional*, defaults to 25 DEFAULT: 25
sigmas	Custom sigmas to use for the denoising process with schedulers which support a sigmas argument in their set_timesteps method. If not defined, the default behavior when num_inference_steps is passed will be used. TYPE: `List[float]`, *optional* DEFAULT: None
min_guidance_scale	The minimum guidance scale. Used for the classifier free guidance with first frame. TYPE: `float`, *optional*, defaults to 1.0 DEFAULT: 1.0
max_guidance_scale	The maximum guidance scale. Used for the classifier free guidance with last frame. TYPE: `float`, *optional*, defaults to 3.0 DEFAULT: 3.0
fps	Frames per second. The rate at which the generated images shall be exported to a video after generation. Note that Stable Diffusion Video's UNet was micro-conditioned on fps-1 during training. TYPE: `int`, *optional*, defaults to 7 DEFAULT: 7
motion_bucket_id	Used for conditioning the amount of motion for the generation. The higher the number the more motion will be in the video. TYPE: `int`, *optional*, defaults to 127 DEFAULT: 127
noise_aug_strength	The amount of noise added to the init image, the higher it is the less the video will look like the init image. Increase it for more motion. TYPE: `float`, *optional*, defaults to 0.02 DEFAULT: 0.02
decode_chunk_size	The number of frames to decode at a time. Higher chunk size leads to better temporal consistency at the expense of more memory usage. By default, the decoder decodes all frames at once for maximal quality. For lower memory usage, reduce decode_chunk_size. TYPE: `int`, *optional* DEFAULT: None
num_videos_per_prompt	The number of videos to generate per prompt. TYPE: `int`, *optional*, defaults to 1 DEFAULT: 1
generator	A np.random.Generator to make generation deterministic. TYPE: `np.random.Generator` or `List[np.random.Generator]`, *optional* DEFAULT: None
latents	Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for video generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied random generator. TYPE: `ms.Tensor`, *optional* DEFAULT: None
output_type	The output format of the generated image. Choose between pil, np or ms. TYPE: `str`, *optional*, defaults to `"pil"` DEFAULT: 'pil'
callback_on_step_end	A function that is called at the end of each denoising step during inference. The function is called with the following arguments: callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict). callback_kwargs will include a list of all tensors as specified by callback_on_step_end_tensor_inputs. TYPE: `Callable`, *optional* DEFAULT: None
callback_on_step_end_tensor_inputs	The list of tensor inputs for the callback_on_step_end function. The tensors specified in the list will be passed as callback_kwargs argument. You will only be able to include variables listed in the ._callback_tensor_inputs attribute of your pipeline class. TYPE: `List`, *optional* DEFAULT: ['latents']
return_dict	Whether or not to return a [~pipelines.stable_diffusion.StableDiffusionPipelineOutput] instead of a plain tuple. TYPE: `bool`, *optional*, defaults to `False` DEFAULT: False

RETURNS	DESCRIPTION
	[~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput] or tuple: If return_dict is True, [~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput] is returned, otherwise a tuple of (List[List[PIL.Image.Image]] or np.ndarray or ms.Tensor) is returned.

Source code in mindone/diffusers/pipelines/stable_video_diffusion/pipeline_stable_video_diffusion.py

def __call__(
    self,
    image: Union[PIL.Image.Image, List[PIL.Image.Image], ms.Tensor],
    height: int = 576,
    width: int = 1024,
    num_frames: Optional[int] = None,
    num_inference_steps: int = 25,
    sigmas: Optional[List[float]] = None,
    min_guidance_scale: float = 1.0,
    max_guidance_scale: float = 3.0,
    fps: int = 7,
    motion_bucket_id: int = 127,
    noise_aug_strength: float = 0.02,
    decode_chunk_size: Optional[int] = None,
    num_videos_per_prompt: Optional[int] = 1,
    generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
    latents: Optional[ms.Tensor] = None,
    output_type: Optional[str] = "pil",
    callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
    callback_on_step_end_tensor_inputs: List[str] = ["latents"],
    return_dict: bool = False,
):
    r"""
    The call function to the pipeline for generation.

    Args:
        image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `ms.Tensor`):
            Image(s) to guide image generation. If you provide a tensor, the expected value range is between `[0, 1]`.
        height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
            The height in pixels of the generated image.
        width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
            The width in pixels of the generated image.
        num_frames (`int`, *optional*):
            The number of video frames to generate. Defaults to `self.unet.config.num_frames`
            (14 for `stable-video-diffusion-img2vid` and to 25 for `stable-video-diffusion-img2vid-xt`).
        num_inference_steps (`int`, *optional*, defaults to 25):
            The number of denoising steps. More denoising steps usually lead to a higher quality video at the
            expense of slower inference. This parameter is modulated by `strength`.
        sigmas (`List[float]`, *optional*):
            Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
            their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
            will be used.
        min_guidance_scale (`float`, *optional*, defaults to 1.0):
            The minimum guidance scale. Used for the classifier free guidance with first frame.
        max_guidance_scale (`float`, *optional*, defaults to 3.0):
            The maximum guidance scale. Used for the classifier free guidance with last frame.
        fps (`int`, *optional*, defaults to 7):
            Frames per second. The rate at which the generated images shall be exported to a video after generation.
            Note that Stable Diffusion Video's UNet was micro-conditioned on fps-1 during training.
        motion_bucket_id (`int`, *optional*, defaults to 127):
            Used for conditioning the amount of motion for the generation. The higher the number the more motion
            will be in the video.
        noise_aug_strength (`float`, *optional*, defaults to 0.02):
            The amount of noise added to the init image, the higher it is the less the video will look like the init image. Increase it for more motion.
        decode_chunk_size (`int`, *optional*):
            The number of frames to decode at a time. Higher chunk size leads to better temporal consistency at the expense of more memory usage.
            By default, the decoder decodes all frames at once for maximal quality. For lower memory usage, reduce `decode_chunk_size`.
        num_videos_per_prompt (`int`, *optional*, defaults to 1):
            The number of videos to generate per prompt.
        generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
            A [`np.random.Generator`](https://numpy.org/doc/stable/reference/random/generator.html) to make
            generation deterministic.
        latents (`ms.Tensor`, *optional*):
            Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for video
            generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
            tensor is generated by sampling using the supplied random `generator`.
        output_type (`str`, *optional*, defaults to `"pil"`):
            The output format of the generated image. Choose between `pil`, `np` or `ms`.
        callback_on_step_end (`Callable`, *optional*):
            A function that is called at the end of each denoising step during inference. The function is called
            with the following arguments:
                `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`.
            `callback_kwargs` will include a list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
        callback_on_step_end_tensor_inputs (`List`, *optional*):
            The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
            will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
            `._callback_tensor_inputs` attribute of your pipeline class.
        return_dict (`bool`, *optional*, defaults to `False`):
            Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
            plain tuple.

    Examples:

    Returns:
        [`~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput`] or `tuple`:
            If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableVideoDiffusionPipelineOutput`] is returned,
            otherwise a `tuple` of (`List[List[PIL.Image.Image]]` or `np.ndarray` or `ms.Tensor`) is returned.
    """
    # 0. Default height and width to unet
    height = height or self.unet.config.sample_size * self.vae_scale_factor
    width = width or self.unet.config.sample_size * self.vae_scale_factor

    num_frames = num_frames if num_frames is not None else self.unet.config.num_frames
    decode_chunk_size = decode_chunk_size if decode_chunk_size is not None else num_frames

    # 1. Check inputs. Raise error if not correct
    self.check_inputs(image, height, width)

    # 2. Define call parameters
    if isinstance(image, PIL.Image.Image):
        batch_size = 1
    elif isinstance(image, list):
        batch_size = len(image)
    else:
        batch_size = image.shape[0]
    # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
    # corresponds to doing no classifier free guidance.
    self._guidance_scale = max_guidance_scale

    # 3. Encode input image
    image_embeddings = self._encode_image(image, num_videos_per_prompt, self.do_classifier_free_guidance)

    # NOTE: Stable Video Diffusion was conditioned on fps - 1, which is why it is reduced here.
    # See: https://github.com/Stability-AI/generative-models/blob/ed0997173f98eaf8f4edf7ba5fe8f15c6b877fd3/scripts/sampling/simple_video_sample.py#L188
    fps = fps - 1

    # 4. Encode input image using VAE
    image = self.video_processor.preprocess(image, height=height, width=width)
    noise = randn_tensor(image.shape, generator=generator, dtype=image.dtype)
    image = image + noise_aug_strength * noise

    needs_upcasting = self.vae.dtype == ms.float16 and self.vae.config.force_upcast
    if needs_upcasting:
        self.vae.to(dtype=ms.float32)

    image_latents = self._encode_vae_image(
        image,
        num_videos_per_prompt=num_videos_per_prompt,
        do_classifier_free_guidance=self.do_classifier_free_guidance,
    )
    image_latents = image_latents.to(image_embeddings.dtype)

    # cast back to fp16 if needed
    if needs_upcasting:
        self.vae.to(dtype=ms.float16)

    # Repeat the image latents for each frame so we can concatenate them with the noise
    # image_latents [batch, channels, height, width] ->[batch, num_frames, channels, height, width]
    image_latents = image_latents.unsqueeze(1).tile((1, num_frames, 1, 1, 1))

    # 5. Get Added Time IDs
    added_time_ids = self._get_add_time_ids(
        fps,
        motion_bucket_id,
        noise_aug_strength,
        image_embeddings.dtype,
        batch_size,
        num_videos_per_prompt,
        self.do_classifier_free_guidance,
    )

    # 6. Prepare timesteps
    timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, None, sigmas)

    # 7. Prepare latent variables
    num_channels_latents = self.unet.config.in_channels
    latents = self.prepare_latents(
        batch_size * num_videos_per_prompt,
        num_frames,
        num_channels_latents,
        height,
        width,
        image_embeddings.dtype,
        generator,
        latents,
    )

    # 8. Prepare guidance scale
    guidance_scale = ms.Tensor.from_numpy(
        np.linspace(min_guidance_scale, max_guidance_scale, num_frames)
    ).unsqueeze(0)
    guidance_scale = guidance_scale.to(latents.dtype)
    guidance_scale = guidance_scale.tile((batch_size * num_videos_per_prompt, 1))
    guidance_scale = _append_dims(guidance_scale, latents.ndim)

    self._guidance_scale = guidance_scale

    # 9. Denoising loop
    num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
    self._num_timesteps = len(timesteps)
    with self.progress_bar(total=num_inference_steps) as progress_bar:
        for i, t in enumerate(timesteps):
            # expand the latents if we are doing classifier free guidance
            latent_model_input = ops.cat([latents] * 2) if self.do_classifier_free_guidance else latents
            latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

            # Concatenate image_latents over channels dimension
            latent_model_input = ops.cat([latent_model_input, image_latents], axis=2)

            # predict the noise residual
            noise_pred = self.unet(
                latent_model_input,
                t,
                encoder_hidden_states=image_embeddings,
                added_time_ids=added_time_ids,
                return_dict=False,
            )[0]

            # perform guidance
            if self.do_classifier_free_guidance:
                noise_pred_uncond, noise_pred_cond = noise_pred.chunk(2)
                noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_cond - noise_pred_uncond)

            # compute the previous noisy sample x_t -> x_t-1
            latents = self.scheduler.step(noise_pred, t, latents)[0]

            if callback_on_step_end is not None:
                callback_kwargs = {}
                for k in callback_on_step_end_tensor_inputs:
                    callback_kwargs[k] = locals()[k]
                callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)

                latents = callback_outputs.pop("latents", latents)

            if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                progress_bar.update()

    if not output_type == "latent":
        # cast back to fp16 if needed
        if needs_upcasting:
            self.vae.to(dtype=ms.float16)
        frames = self.decode_latents(latents, num_frames, decode_chunk_size)
        frames = self.video_processor.postprocess_video(video=frames, output_type=output_type)
    else:
        frames = latents

    if not return_dict:
        return frames

    return StableVideoDiffusionPipelineOutput(frames=frames)

mindone.diffusers.pipelines.stable_video_diffusion.StableVideoDiffusionPipelineOutput dataclass

Bases: BaseOutput

Output class for Stable Video Diffusion pipeline.

PARAMETER	DESCRIPTION
frames	List of denoised PIL images of length batch_size or numpy array or ms tensor of shape (batch_size, num_frames, height, width, num_channels). TYPE: `[List[List[PIL.Image.Image]]`, `np.ndarray`, `ms.Tensor`]

Source code in mindone/diffusers/pipelines/stable_video_diffusion/pipeline_stable_video_diffusion.py

@dataclass
class StableVideoDiffusionPipelineOutput(BaseOutput):
    r"""
    Output class for Stable Video Diffusion pipeline.

    Args:
        frames (`[List[List[PIL.Image.Image]]`, `np.ndarray`, `ms.Tensor`]):
            List of denoised PIL images of length `batch_size` or numpy array or ms tensor of shape `(batch_size,
            num_frames, height, width, num_channels)`.
    """

    frames: Union[List[List[PIL.Image.Image]], np.ndarray, ms.Tensor]