I2VGen-XL

I2VGen-XL: High-Quality Image-to-Video Synthesis via Cascaded Diffusion Models by Shiwei Zhang, Jiayu Wang, Yingya Zhang, Kang Zhao, Hangjie Yuan, Zhiwu Qin, Xiang Wang, Deli Zhao, and Jingren Zhou.

The abstract from the paper is:

Video synthesis has recently made remarkable strides benefiting from the rapid development of diffusion models. However, it still encounters challenges in terms of semantic accuracy, clarity and spatio-temporal continuity. They primarily arise from the scarcity of well-aligned text-video data and the complex inherent structure of videos, making it difficult for the model to simultaneously ensure semantic and qualitative excellence. In this report, we propose a cascaded I2VGen-XL approach that enhances model performance by decoupling these two factors and ensures the alignment of the input data by utilizing static images as a form of crucial guidance. I2VGen-XL consists of two stages: i) the base stage guarantees coherent semantics and preserves content from input images by using two hierarchical encoders, and ii) the refinement stage enhances the video's details by incorporating an additional brief text and improves the resolution to 1280×720. To improve the diversity, we collect around 35 million single-shot text-video pairs and 6 billion text-image pairs to optimize the model. By this means, I2VGen-XL can simultaneously enhance the semantic accuracy, continuity of details and clarity of generated videos. Through extensive experiments, we have investigated the underlying principles of I2VGen-XL and compared it with current top methods, which can demonstrate its effectiveness on diverse data. The source code and models will be publicly available at this https URL.

The original codebase can be found here. The model checkpoints can be found here.

Tip

Make sure to check out the Schedulers guide to learn how to explore the tradeoff between scheduler speed and quality, and see the reuse components across pipelines section to learn how to efficiently load the same components into multiple pipelines. Also, to know more about reducing the memory usage of this pipeline, refer to the ["Reduce memory usage"] section here.

Sample output with I2VGenXL:

library.

Notes

• I2VGenXL always uses a clip_skip value of 1. This means it leverages the penultimate layer representations from the text encoder of CLIP.
• It can generate videos of quality that is often on par with Stable Video Diffusion (SVD).
• Unlike SVD, it additionally accepts text prompts as inputs.
• It can generate higher resolution videos.
• When using the DDIMScheduler (which is default for this pipeline), less than 50 steps for inference leads to bad results.
• This implementation is 1-stage variant of I2VGenXL. The main figure in the I2VGen-XL paper shows a 2-stage variant, however, 1-stage variant works well. See this discussion for more details.

mindone.diffusers.I2VGenXLPipeline

Bases: DiffusionPipeline, StableDiffusionMixin

Pipeline for image-to-video generation as proposed in I2VGenXL.

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: [`AutoencoderKL`]
text_encoder	Frozen text-encoder (clip-vit-large-patch14). TYPE: [`CLIPTextModel`]
tokenizer	A [~transformers.CLIPTokenizer] to tokenize text. TYPE: `CLIPTokenizer`
unet	A [I2VGenXLUNet] to denoise the encoded video latents. TYPE: [`I2VGenXLUNet`]
scheduler	A scheduler to be used in combination with unet to denoise the encoded image latents. TYPE: [`DDIMScheduler`]

Source code in mindone/diffusers/pipelines/i2vgen_xl/pipeline_i2vgen_xl.py

class I2VGenXLPipeline(DiffusionPipeline, StableDiffusionMixin):
    r"""
    Pipeline for image-to-video generation as proposed in [I2VGenXL](https://i2vgen-xl.github.io/).

    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 ([`AutoencoderKL`]):
            Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
        text_encoder ([`CLIPTextModel`]):
            Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
        tokenizer (`CLIPTokenizer`):
            A [`~transformers.CLIPTokenizer`] to tokenize text.
        unet ([`I2VGenXLUNet`]):
            A [`I2VGenXLUNet`] to denoise the encoded video latents.
        scheduler ([`DDIMScheduler`]):
            A scheduler to be used in combination with `unet` to denoise the encoded image latents.
    """

    model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"

    def __init__(
        self,
        vae: AutoencoderKL,
        text_encoder: CLIPTextModel,
        tokenizer: CLIPTokenizer,
        image_encoder: CLIPVisionModelWithProjection,
        feature_extractor: CLIPImageProcessor,
        unet: I2VGenXLUNet,
        scheduler: DDIMScheduler,
    ):
        super().__init__()

        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            image_encoder=image_encoder,
            feature_extractor=feature_extractor,
            unet=unet,
            scheduler=scheduler,
        )
        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
        # `do_resize=False` as we do custom resizing.
        self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor, do_resize=False)

    @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):
        return self._guidance_scale > 1

    def encode_prompt(
        self,
        prompt,
        num_videos_per_prompt,
        negative_prompt=None,
        prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        clip_skip: Optional[int] = None,
    ):
        r"""
        Encodes the prompt into text encoder hidden states.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            num_videos_per_prompt (`int`):
                number of images that should be generated per prompt
            do_classifier_free_guidance (`bool`):
                whether to use classifier free guidance or not
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts not to guide the image generation. If not defined, one has to pass
                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
                less than `1`).
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
            negative_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
                argument.
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.
        """
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        if prompt_embeds is None:
            text_inputs = self.tokenizer(
                prompt,
                padding="max_length",
                max_length=self.tokenizer.model_max_length,
                truncation=True,
                return_tensors="np",
            )
            text_input_ids = text_inputs.input_ids
            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="np").input_ids

            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not np.array_equal(
                text_input_ids, untruncated_ids
            ):
                removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1])
                logger.warning(
                    "The following part of your input was truncated because CLIP can only handle sequences up to"
                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
                )

            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
                attention_mask = text_inputs.attention_mask
            else:
                attention_mask = None

            if clip_skip is None:
                prompt_embeds = self.text_encoder(ms.tensor(text_input_ids), attention_mask=attention_mask)
                prompt_embeds = prompt_embeds[0]
            else:
                prompt_embeds = self.text_encoder(
                    ms.tensor(text_input_ids), attention_mask=attention_mask, output_hidden_states=True
                )
                # Access the `hidden_states` first, that contains a tuple of
                # all the hidden states from the encoder layers. Then index into
                # the tuple to access the hidden states from the desired layer.
                prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
                # We also need to apply the final LayerNorm here to not mess with the
                # representations. The `last_hidden_states` that we typically use for
                # obtaining the final prompt representations passes through the LayerNorm
                # layer.
                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)

        if self.text_encoder is not None:
            prompt_embeds_dtype = self.text_encoder.dtype
        elif self.unet is not None:
            prompt_embeds_dtype = self.unet.dtype
        else:
            prompt_embeds_dtype = prompt_embeds.dtype

        prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype)

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

        # get unconditional embeddings for classifier free guidance
        if self.do_classifier_free_guidance and negative_prompt_embeds is None:
            uncond_tokens: List[str]
            if negative_prompt is None:
                uncond_tokens = [""] * batch_size
            elif prompt is not None and type(prompt) is not type(negative_prompt):
                raise TypeError(
                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                    f" {type(prompt)}."
                )
            elif isinstance(negative_prompt, str):
                uncond_tokens = [negative_prompt]
            elif batch_size != len(negative_prompt):
                raise ValueError(
                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                    " the batch size of `prompt`."
                )
            else:
                uncond_tokens = negative_prompt

            max_length = prompt_embeds.shape[1]
            uncond_input = self.tokenizer(
                uncond_tokens,
                padding="max_length",
                max_length=max_length,
                truncation=True,
                return_tensors="np",
            )

            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
                attention_mask = ms.Tensor.from_numpy(uncond_input.attention_mask)
            else:
                attention_mask = None

            # Apply clip_skip to negative prompt embeds
            if clip_skip is None:
                negative_prompt_embeds = self.text_encoder(
                    ms.Tensor.from_numpy(uncond_input.input_ids),
                    attention_mask=attention_mask,
                )
                negative_prompt_embeds = negative_prompt_embeds[0]
            else:
                negative_prompt_embeds = self.text_encoder(
                    ms.Tensor.from_numpy(uncond_input.input_ids),
                    attention_mask=attention_mask,
                    output_hidden_states=True,
                )
                # Access the `hidden_states` first, that contains a tuple of
                # all the hidden states from the encoder layers. Then index into
                # the tuple to access the hidden states from the desired layer.
                negative_prompt_embeds = negative_prompt_embeds[-1][-(clip_skip + 1)]
                # We also need to apply the final LayerNorm here to not mess with the
                # representations. The `last_hidden_states` that we typically use for
                # obtaining the final prompt representations passes through the LayerNorm
                # layer.
                negative_prompt_embeds = self.text_encoder.text_model.final_layer_norm(negative_prompt_embeds)

        if self.do_classifier_free_guidance:
            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
            seq_len = negative_prompt_embeds.shape[1]

            negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype)

            negative_prompt_embeds = negative_prompt_embeds.tile((1, num_videos_per_prompt, 1))
            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)

        return prompt_embeds, negative_prompt_embeds

    def _encode_image(self, image, num_videos_per_prompt):
        dtype = next(self.image_encoder.get_parameters()).dtype

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

            # Normalize the image with CLIP training stats.
            image = self.feature_extractor(
                images=image,
                do_normalize=True,
                do_center_crop=False,
                do_resize=False,
                do_rescale=False,
                return_tensors="np",
            ).pixel_values
            image = ms.Tensor(image)

        image = 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 self.do_classifier_free_guidance:
            negative_image_embeddings = ops.zeros_like(image_embeddings)
            image_embeddings = ops.cat([negative_image_embeddings, image_embeddings])

        return image_embeddings

    def decode_latents(self, latents, decode_chunk_size=None):
        latents = 1 / self.vae.config.scaling_factor * latents

        batch_size, channels, num_frames, height, width = latents.shape
        latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * num_frames, channels, height, width)

        if decode_chunk_size is not None:
            frames = []
            for i in range(0, latents.shape[0], decode_chunk_size):
                frame = self.vae.decode(latents[i : i + decode_chunk_size])[0]
                frames.append(frame)
            image = ops.cat(frames, axis=0)
        else:
            image = self.vae.decode(latents)[0]

        decode_shape = (batch_size, num_frames, -1) + image.shape[2:]
        video = image[None, :].reshape(decode_shape).permute(0, 2, 1, 3, 4)

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

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
    def prepare_extra_step_kwargs(self, generator, eta):
        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
        # and should be between [0, 1]

        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
        extra_step_kwargs = {}
        if accepts_eta:
            extra_step_kwargs["eta"] = eta

        # check if the scheduler accepts generator
        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
        if accepts_generator:
            extra_step_kwargs["generator"] = generator
        return extra_step_kwargs

    def check_inputs(
        self,
        prompt,
        image,
        height,
        width,
        negative_prompt=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
    ):
        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}.")

        if prompt is not None and prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
                " only forward one of the two."
            )
        elif prompt is None and prompt_embeds is None:
            raise ValueError(
                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
            )
        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")

        if negative_prompt is not None and negative_prompt_embeds is not None:
            raise ValueError(
                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
            )

        if prompt_embeds is not None and negative_prompt_embeds is not None:
            if prompt_embeds.shape != negative_prompt_embeds.shape:
                raise ValueError(
                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
                    f" {negative_prompt_embeds.shape}."
                )

        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)}"
            )

    def prepare_image_latents(
        self,
        image,
        num_frames,
        num_videos_per_prompt,
    ):
        image_latents = self.vae.encode(image)[0]
        image_latents = self.vae.diag_gauss_dist.sample(image_latents)
        image_latents = image_latents * self.vae.config.scaling_factor

        # Add frames dimension to image latents
        image_latents = image_latents.unsqueeze(2)

        # Append a position mask for each subsequent frame
        # after the intial image latent frame
        frame_position_mask = []
        for frame_idx in range(num_frames - 1):
            scale = (frame_idx + 1) / (num_frames - 1)
            frame_position_mask.append(ops.ones_like(image_latents[:, :, :1]) * scale)
        if frame_position_mask:
            frame_position_mask = ops.cat(frame_position_mask, axis=2)
            image_latents = ops.cat([image_latents, frame_position_mask], axis=2)

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

        if self.do_classifier_free_guidance:
            image_latents = ops.cat([image_latents] * 2)

        return image_latents

    # Copied from diffusers.pipelines.text_to_video_synthesis.pipeline_text_to_video_synth.TextToVideoSDPipeline.prepare_latents
    def prepare_latents(
        self, batch_size, num_channels_latents, num_frames, height, width, dtype, generator, latents=None
    ):
        shape = (
            batch_size,
            num_channels_latents,
            num_frames,
            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

    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        image: PipelineImageInput = None,
        height: Optional[int] = 704,
        width: Optional[int] = 1280,
        target_fps: Optional[int] = 16,
        num_frames: int = 16,
        num_inference_steps: int = 50,
        guidance_scale: float = 9.0,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        eta: float = 0.0,
        num_videos_per_prompt: Optional[int] = 1,
        decode_chunk_size: Optional[int] = 1,
        generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
        latents: Optional[ms.Tensor] = None,
        prompt_embeds: Optional[ms.Tensor] = None,
        negative_prompt_embeds: Optional[ms.Tensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = False,
        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
        clip_skip: Optional[int] = 1,
    ):
        r"""
        The call function to the pipeline for image-to-video generation with [`I2VGenXLPipeline`].

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
            image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `ms.Tensor`):
                Image or images to guide image generation. If you provide a tensor, it needs to be compatible with
                [`CLIPImageProcessor`](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json).
            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.
            target_fps (`int`, *optional*):
                Frames per second. The rate at which the generated images shall be exported to a video after
                generation. This is also used as a "micro-condition" while generation.
            num_frames (`int`, *optional*):
                The number of video frames to generate.
            num_inference_steps (`int`, *optional*):
                The number of denoising steps.
            guidance_scale (`float`, *optional*, defaults to 7.5):
                A higher guidance scale value encourages the model to generate images closely linked to the text
                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
            negative_prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
            eta (`float`, *optional*):
                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
            num_videos_per_prompt (`int`, *optional*):
                The number of images to generate per prompt.
            decode_chunk_size (`int`, *optional*):
                The number of frames to decode at a time. The higher the chunk size, the higher the temporal
                consistency between frames, but also the higher the memory consumption. By default, the decoder will
                decode all frames at once for maximal quality. Reduce `decode_chunk_size` to reduce memory usage.
            generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
                A [`np.random.Generator`](https://pytorch.org/docs/stable/generated/torch.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 image
                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`.
            prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
                provided, text embeddings are generated from the `prompt` input argument.
            negative_prompt_embeds (`ms.Tensor`, *optional*):
                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
            output_type (`str`, *optional*, defaults to `"pil"`):
                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
            return_dict (`bool`, *optional*, defaults to `False`):
                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
                plain tuple.
            cross_attention_kwargs (`dict`, *optional*):
                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
            clip_skip (`int`, *optional*):
                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                the output of the pre-final layer will be used for computing the prompt embeddings.

        Examples:

        Returns:
            [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] or `tuple`:
                If `return_dict` is `True`, [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] is
                returned, otherwise a `tuple` is returned where the first element is a list with the generated frames.
        """
        # Warning if self.unet datatype is float32
        if self.unet.dtype == ms.float32:
            logger.warning(
                "\033[93m[WARNING] \033[0m"
                "The dtype of I2VGenXLUNet is mindspore.float32, which could result in unexpected output because the nn.Conv3d layer "
                "used in unet is not fully supported in float32 on Ascend. Please consider using lower precision such as float16."
            )

        # 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

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

        # 2. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.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 = guidance_scale

        # 3.1 Encode input text prompt
        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
            prompt,
            num_videos_per_prompt,
            negative_prompt,
            prompt_embeds=prompt_embeds,
            negative_prompt_embeds=negative_prompt_embeds,
            clip_skip=clip_skip,
        )
        # 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
        if self.do_classifier_free_guidance:
            prompt_embeds = ops.cat([negative_prompt_embeds, prompt_embeds])

        # 3.2 Encode image prompt
        # 3.2.1 Image encodings.
        # https://github.com/ali-vilab/i2vgen-xl/blob/2539c9262ff8a2a22fa9daecbfd13f0a2dbc32d0/tools/inferences/inference_i2vgen_entrance.py#L114
        cropped_image = _center_crop_wide(image, (width, width))
        cropped_image = _resize_bilinear(
            cropped_image, (self.feature_extractor.crop_size["width"], self.feature_extractor.crop_size["height"])
        )
        image_embeddings = self._encode_image(cropped_image, num_videos_per_prompt)

        # 3.2.2 Image latents.
        resized_image = _center_crop_wide(image, (width, height))
        image = self.video_processor.preprocess(resized_image).to(dtype=image_embeddings.dtype)
        image_latents = self.prepare_image_latents(
            image,
            num_frames=num_frames,
            num_videos_per_prompt=num_videos_per_prompt,
        )

        # 3.3 Prepare additional conditions for the UNet.
        if self.do_classifier_free_guidance:
            fps_tensor = ms.Tensor([target_fps, target_fps])
        else:
            fps_tensor = ms.Tensor([target_fps])
        fps_tensor = fps_tensor.tile((batch_size * num_videos_per_prompt, 1)).ravel()

        # 4. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps)
        timesteps = self.scheduler.timesteps

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

        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

        # 7. Denoising loop
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
        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
                # TODO: method of scheduler should not change the dtype of input.
                #  Remove the casting after cuiyushi confirm that.
                tmp_dtype = latent_model_input.dtype
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
                latent_model_input = latent_model_input.to(tmp_dtype)

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=prompt_embeds,
                    fps=fps_tensor,
                    image_latents=image_latents,
                    image_embeddings=image_embeddings,
                    cross_attention_kwargs=cross_attention_kwargs,
                    return_dict=False,
                )[0]

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

                # reshape latents
                batch_size, channel, frames, width, height = latents.shape
                latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)
                noise_pred = noise_pred.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)

                # compute the previous noisy sample x_t -> x_t-1
                # TODO: method of scheduler should not change the dtype of input.
                #  Remove the casting after cuiyushi confirm that.
                tmp_dtype = latents.dtype
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)[0]
                latents = latents.to(tmp_dtype)

                # reshape latents back
                latents = latents[None, :].reshape(batch_size, frames, channel, width, height).permute(0, 2, 1, 3, 4)
                # call the callback, if provided
                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                    progress_bar.update()

        # 8. Post processing
        if output_type == "latent":
            video = latents
        else:
            video_tensor = self.decode_latents(latents, decode_chunk_size=decode_chunk_size)
            video = self.video_processor.postprocess_video(video=video_tensor, output_type=output_type)

        if not return_dict:
            return (video,)

        return I2VGenXLPipelineOutput(frames=video)

mindone.diffusers.I2VGenXLPipeline.__call__(prompt=None, image=None, height=704, width=1280, target_fps=16, num_frames=16, num_inference_steps=50, guidance_scale=9.0, negative_prompt=None, eta=0.0, num_videos_per_prompt=1, decode_chunk_size=1, generator=None, latents=None, prompt_embeds=None, negative_prompt_embeds=None, output_type='pil', return_dict=False, cross_attention_kwargs=None, clip_skip=1)

The call function to the pipeline for image-to-video generation with [I2VGenXLPipeline].

PARAMETER	DESCRIPTION
prompt	The prompt or prompts to guide image generation. If not defined, you need to pass prompt_embeds. TYPE: `str` or `List[str]`, *optional* DEFAULT: None
image	Image or images to guide image generation. If you provide a tensor, it needs to be compatible with CLIPImageProcessor. TYPE: `PIL.Image.Image` or `List[PIL.Image.Image]` or `ms.Tensor` DEFAULT: None
height	The height in pixels of the generated image. TYPE: `int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor` DEFAULT: 704
width	The width in pixels of the generated image. TYPE: `int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor` DEFAULT: 1280
target_fps	Frames per second. The rate at which the generated images shall be exported to a video after generation. This is also used as a "micro-condition" while generation. TYPE: `int`, *optional* DEFAULT: 16
num_frames	The number of video frames to generate. TYPE: `int`, *optional* DEFAULT: 16
num_inference_steps	The number of denoising steps. TYPE: `int`, *optional* DEFAULT: 50
guidance_scale	A higher guidance scale value encourages the model to generate images closely linked to the text prompt at the expense of lower image quality. Guidance scale is enabled when guidance_scale > 1. TYPE: `float`, *optional*, defaults to 7.5 DEFAULT: 9.0
negative_prompt	The prompt or prompts to guide what to not include in image generation. If not defined, you need to pass negative_prompt_embeds instead. Ignored when not using guidance (guidance_scale < 1). TYPE: `str` or `List[str]`, *optional* DEFAULT: None
eta	Corresponds to parameter eta (η) from the DDIM paper. Only applies to the [~schedulers.DDIMScheduler], and is ignored in other schedulers. TYPE: `float`, *optional* DEFAULT: 0.0
num_videos_per_prompt	The number of images to generate per prompt. TYPE: `int`, *optional* DEFAULT: 1
decode_chunk_size	The number of frames to decode at a time. The higher the chunk size, the higher the temporal consistency between frames, but also the higher the memory consumption. By default, the decoder will decode all frames at once for maximal quality. Reduce decode_chunk_size to reduce memory usage. TYPE: `int`, *optional* 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 image 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
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from the prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
negative_prompt_embeds	Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, negative_prompt_embeds are generated from the negative_prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
output_type	The output format of the generated image. Choose between PIL.Image or np.array. TYPE: `str`, *optional*, defaults to `"pil"` DEFAULT: 'pil'
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
cross_attention_kwargs	A kwargs dictionary that if specified is passed along to the [AttentionProcessor] as defined in self.processor. TYPE: `dict`, *optional* DEFAULT: None
clip_skip	Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that the output of the pre-final layer will be used for computing the prompt embeddings. TYPE: `int`, *optional* DEFAULT: 1

RETURNS	DESCRIPTION
	[pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput] or tuple: If return_dict is True, [pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput] is returned, otherwise a tuple is returned where the first element is a list with the generated frames.

Source code in mindone/diffusers/pipelines/i2vgen_xl/pipeline_i2vgen_xl.py

def __call__(
    self,
    prompt: Union[str, List[str]] = None,
    image: PipelineImageInput = None,
    height: Optional[int] = 704,
    width: Optional[int] = 1280,
    target_fps: Optional[int] = 16,
    num_frames: int = 16,
    num_inference_steps: int = 50,
    guidance_scale: float = 9.0,
    negative_prompt: Optional[Union[str, List[str]]] = None,
    eta: float = 0.0,
    num_videos_per_prompt: Optional[int] = 1,
    decode_chunk_size: Optional[int] = 1,
    generator: Optional[Union[np.random.Generator, List[np.random.Generator]]] = None,
    latents: Optional[ms.Tensor] = None,
    prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    output_type: Optional[str] = "pil",
    return_dict: bool = False,
    cross_attention_kwargs: Optional[Dict[str, Any]] = None,
    clip_skip: Optional[int] = 1,
):
    r"""
    The call function to the pipeline for image-to-video generation with [`I2VGenXLPipeline`].

    Args:
        prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
        image (`PIL.Image.Image` or `List[PIL.Image.Image]` or `ms.Tensor`):
            Image or images to guide image generation. If you provide a tensor, it needs to be compatible with
            [`CLIPImageProcessor`](https://huggingface.co/lambdalabs/sd-image-variations-diffusers/blob/main/feature_extractor/preprocessor_config.json).
        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.
        target_fps (`int`, *optional*):
            Frames per second. The rate at which the generated images shall be exported to a video after
            generation. This is also used as a "micro-condition" while generation.
        num_frames (`int`, *optional*):
            The number of video frames to generate.
        num_inference_steps (`int`, *optional*):
            The number of denoising steps.
        guidance_scale (`float`, *optional*, defaults to 7.5):
            A higher guidance scale value encourages the model to generate images closely linked to the text
            `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
        negative_prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts to guide what to not include in image generation. If not defined, you need to
            pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
        eta (`float`, *optional*):
            Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
            to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
        num_videos_per_prompt (`int`, *optional*):
            The number of images to generate per prompt.
        decode_chunk_size (`int`, *optional*):
            The number of frames to decode at a time. The higher the chunk size, the higher the temporal
            consistency between frames, but also the higher the memory consumption. By default, the decoder will
            decode all frames at once for maximal quality. Reduce `decode_chunk_size` to reduce memory usage.
        generator (`np.random.Generator` or `List[np.random.Generator]`, *optional*):
            A [`np.random.Generator`](https://pytorch.org/docs/stable/generated/torch.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 image
            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`.
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
            provided, text embeddings are generated from the `prompt` input argument.
        negative_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
            not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
        output_type (`str`, *optional*, defaults to `"pil"`):
            The output format of the generated image. Choose between `PIL.Image` or `np.array`.
        return_dict (`bool`, *optional*, defaults to `False`):
            Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
            plain tuple.
        cross_attention_kwargs (`dict`, *optional*):
            A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
            [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
        clip_skip (`int`, *optional*):
            Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
            the output of the pre-final layer will be used for computing the prompt embeddings.

    Examples:

    Returns:
        [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] or `tuple`:
            If `return_dict` is `True`, [`pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput`] is
            returned, otherwise a `tuple` is returned where the first element is a list with the generated frames.
    """
    # Warning if self.unet datatype is float32
    if self.unet.dtype == ms.float32:
        logger.warning(
            "\033[93m[WARNING] \033[0m"
            "The dtype of I2VGenXLUNet is mindspore.float32, which could result in unexpected output because the nn.Conv3d layer "
            "used in unet is not fully supported in float32 on Ascend. Please consider using lower precision such as float16."
        )

    # 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

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

    # 2. Define call parameters
    if prompt is not None and isinstance(prompt, str):
        batch_size = 1
    elif prompt is not None and isinstance(prompt, list):
        batch_size = len(prompt)
    else:
        batch_size = prompt_embeds.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 = guidance_scale

    # 3.1 Encode input text prompt
    prompt_embeds, negative_prompt_embeds = self.encode_prompt(
        prompt,
        num_videos_per_prompt,
        negative_prompt,
        prompt_embeds=prompt_embeds,
        negative_prompt_embeds=negative_prompt_embeds,
        clip_skip=clip_skip,
    )
    # 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
    if self.do_classifier_free_guidance:
        prompt_embeds = ops.cat([negative_prompt_embeds, prompt_embeds])

    # 3.2 Encode image prompt
    # 3.2.1 Image encodings.
    # https://github.com/ali-vilab/i2vgen-xl/blob/2539c9262ff8a2a22fa9daecbfd13f0a2dbc32d0/tools/inferences/inference_i2vgen_entrance.py#L114
    cropped_image = _center_crop_wide(image, (width, width))
    cropped_image = _resize_bilinear(
        cropped_image, (self.feature_extractor.crop_size["width"], self.feature_extractor.crop_size["height"])
    )
    image_embeddings = self._encode_image(cropped_image, num_videos_per_prompt)

    # 3.2.2 Image latents.
    resized_image = _center_crop_wide(image, (width, height))
    image = self.video_processor.preprocess(resized_image).to(dtype=image_embeddings.dtype)
    image_latents = self.prepare_image_latents(
        image,
        num_frames=num_frames,
        num_videos_per_prompt=num_videos_per_prompt,
    )

    # 3.3 Prepare additional conditions for the UNet.
    if self.do_classifier_free_guidance:
        fps_tensor = ms.Tensor([target_fps, target_fps])
    else:
        fps_tensor = ms.Tensor([target_fps])
    fps_tensor = fps_tensor.tile((batch_size * num_videos_per_prompt, 1)).ravel()

    # 4. Prepare timesteps
    self.scheduler.set_timesteps(num_inference_steps)
    timesteps = self.scheduler.timesteps

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

    # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
    extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)

    # 7. Denoising loop
    num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
    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
            # TODO: method of scheduler should not change the dtype of input.
            #  Remove the casting after cuiyushi confirm that.
            tmp_dtype = latent_model_input.dtype
            latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
            latent_model_input = latent_model_input.to(tmp_dtype)

            # predict the noise residual
            noise_pred = self.unet(
                latent_model_input,
                t,
                encoder_hidden_states=prompt_embeds,
                fps=fps_tensor,
                image_latents=image_latents,
                image_embeddings=image_embeddings,
                cross_attention_kwargs=cross_attention_kwargs,
                return_dict=False,
            )[0]

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

            # reshape latents
            batch_size, channel, frames, width, height = latents.shape
            latents = latents.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)
            noise_pred = noise_pred.permute(0, 2, 1, 3, 4).reshape(batch_size * frames, channel, width, height)

            # compute the previous noisy sample x_t -> x_t-1
            # TODO: method of scheduler should not change the dtype of input.
            #  Remove the casting after cuiyushi confirm that.
            tmp_dtype = latents.dtype
            latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs)[0]
            latents = latents.to(tmp_dtype)

            # reshape latents back
            latents = latents[None, :].reshape(batch_size, frames, channel, width, height).permute(0, 2, 1, 3, 4)
            # call the callback, if provided
            if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                progress_bar.update()

    # 8. Post processing
    if output_type == "latent":
        video = latents
    else:
        video_tensor = self.decode_latents(latents, decode_chunk_size=decode_chunk_size)
        video = self.video_processor.postprocess_video(video=video_tensor, output_type=output_type)

    if not return_dict:
        return (video,)

    return I2VGenXLPipelineOutput(frames=video)

mindone.diffusers.I2VGenXLPipeline.encode_prompt(prompt, num_videos_per_prompt, negative_prompt=None, prompt_embeds=None, negative_prompt_embeds=None, clip_skip=None)

Encodes the prompt into text encoder hidden states.

PARAMETER	DESCRIPTION
prompt	prompt to be encoded TYPE: `str` or `List[str]`, *optional*
num_videos_per_prompt	number of images that should be generated per prompt TYPE: `int`
do_classifier_free_guidance	whether to use classifier free guidance or not TYPE: `bool`
negative_prompt	The prompt or prompts not to guide the image generation. If not defined, one has to pass negative_prompt_embeds instead. Ignored when not using guidance (i.e., ignored if guidance_scale is less than 1). TYPE: `str` or `List[str]`, *optional* DEFAULT: None
prompt_embeds	Pre-generated text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, text embeddings will be generated from prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
negative_prompt_embeds	Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated from negative_prompt input argument. TYPE: `ms.Tensor`, *optional* DEFAULT: None
clip_skip	Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that the output of the pre-final layer will be used for computing the prompt embeddings. TYPE: `int`, *optional* DEFAULT: None

Source code in mindone/diffusers/pipelines/i2vgen_xl/pipeline_i2vgen_xl.py

def encode_prompt(
    self,
    prompt,
    num_videos_per_prompt,
    negative_prompt=None,
    prompt_embeds: Optional[ms.Tensor] = None,
    negative_prompt_embeds: Optional[ms.Tensor] = None,
    clip_skip: Optional[int] = None,
):
    r"""
    Encodes the prompt into text encoder hidden states.

    Args:
        prompt (`str` or `List[str]`, *optional*):
            prompt to be encoded
        num_videos_per_prompt (`int`):
            number of images that should be generated per prompt
        do_classifier_free_guidance (`bool`):
            whether to use classifier free guidance or not
        negative_prompt (`str` or `List[str]`, *optional*):
            The prompt or prompts not to guide the image generation. If not defined, one has to pass
            `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
            less than `1`).
        prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
            provided, text embeddings will be generated from `prompt` input argument.
        negative_prompt_embeds (`ms.Tensor`, *optional*):
            Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
            weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
            argument.
        clip_skip (`int`, *optional*):
            Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
            the output of the pre-final layer will be used for computing the prompt embeddings.
    """
    if prompt is not None and isinstance(prompt, str):
        batch_size = 1
    elif prompt is not None and isinstance(prompt, list):
        batch_size = len(prompt)
    else:
        batch_size = prompt_embeds.shape[0]

    if prompt_embeds is None:
        text_inputs = self.tokenizer(
            prompt,
            padding="max_length",
            max_length=self.tokenizer.model_max_length,
            truncation=True,
            return_tensors="np",
        )
        text_input_ids = text_inputs.input_ids
        untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="np").input_ids

        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not np.array_equal(
            text_input_ids, untruncated_ids
        ):
            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1])
            logger.warning(
                "The following part of your input was truncated because CLIP can only handle sequences up to"
                f" {self.tokenizer.model_max_length} tokens: {removed_text}"
            )

        if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
            attention_mask = text_inputs.attention_mask
        else:
            attention_mask = None

        if clip_skip is None:
            prompt_embeds = self.text_encoder(ms.tensor(text_input_ids), attention_mask=attention_mask)
            prompt_embeds = prompt_embeds[0]
        else:
            prompt_embeds = self.text_encoder(
                ms.tensor(text_input_ids), attention_mask=attention_mask, output_hidden_states=True
            )
            # Access the `hidden_states` first, that contains a tuple of
            # all the hidden states from the encoder layers. Then index into
            # the tuple to access the hidden states from the desired layer.
            prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
            # We also need to apply the final LayerNorm here to not mess with the
            # representations. The `last_hidden_states` that we typically use for
            # obtaining the final prompt representations passes through the LayerNorm
            # layer.
            prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)

    if self.text_encoder is not None:
        prompt_embeds_dtype = self.text_encoder.dtype
    elif self.unet is not None:
        prompt_embeds_dtype = self.unet.dtype
    else:
        prompt_embeds_dtype = prompt_embeds.dtype

    prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype)

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

    # get unconditional embeddings for classifier free guidance
    if self.do_classifier_free_guidance and negative_prompt_embeds is None:
        uncond_tokens: List[str]
        if negative_prompt is None:
            uncond_tokens = [""] * batch_size
        elif prompt is not None and type(prompt) is not type(negative_prompt):
            raise TypeError(
                f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
                f" {type(prompt)}."
            )
        elif isinstance(negative_prompt, str):
            uncond_tokens = [negative_prompt]
        elif batch_size != len(negative_prompt):
            raise ValueError(
                f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
                f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
                " the batch size of `prompt`."
            )
        else:
            uncond_tokens = negative_prompt

        max_length = prompt_embeds.shape[1]
        uncond_input = self.tokenizer(
            uncond_tokens,
            padding="max_length",
            max_length=max_length,
            truncation=True,
            return_tensors="np",
        )

        if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
            attention_mask = ms.Tensor.from_numpy(uncond_input.attention_mask)
        else:
            attention_mask = None

        # Apply clip_skip to negative prompt embeds
        if clip_skip is None:
            negative_prompt_embeds = self.text_encoder(
                ms.Tensor.from_numpy(uncond_input.input_ids),
                attention_mask=attention_mask,
            )
            negative_prompt_embeds = negative_prompt_embeds[0]
        else:
            negative_prompt_embeds = self.text_encoder(
                ms.Tensor.from_numpy(uncond_input.input_ids),
                attention_mask=attention_mask,
                output_hidden_states=True,
            )
            # Access the `hidden_states` first, that contains a tuple of
            # all the hidden states from the encoder layers. Then index into
            # the tuple to access the hidden states from the desired layer.
            negative_prompt_embeds = negative_prompt_embeds[-1][-(clip_skip + 1)]
            # We also need to apply the final LayerNorm here to not mess with the
            # representations. The `last_hidden_states` that we typically use for
            # obtaining the final prompt representations passes through the LayerNorm
            # layer.
            negative_prompt_embeds = self.text_encoder.text_model.final_layer_norm(negative_prompt_embeds)

    if self.do_classifier_free_guidance:
        # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
        seq_len = negative_prompt_embeds.shape[1]

        negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype)

        negative_prompt_embeds = negative_prompt_embeds.tile((1, num_videos_per_prompt, 1))
        negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)

    return prompt_embeds, negative_prompt_embeds

mindone.diffusers.pipelines.i2vgen_xl.pipeline_i2vgen_xl.I2VGenXLPipelineOutput dataclass

Bases: BaseOutput

Output class for image-to-video pipeline.

Args: frames (torch.Tensor, np.ndarray, or List[List[PIL.Image.Image]]): List of video outputs - It can be a nested list of length batch_size, with each sub-list containing denoised PIL image sequences of length num_frames. It can also be a NumPy array or Torch tensor of shape (batch_size, num_frames, channels, height, width)

Source code in mindone/diffusers/pipelines/i2vgen_xl/pipeline_i2vgen_xl.py

@dataclass
class I2VGenXLPipelineOutput(BaseOutput):
    r"""
     Output class for image-to-video pipeline.

     Args:
        frames (`torch.Tensor`, `np.ndarray`, or List[List[PIL.Image.Image]]):
            List of video outputs - It can be a nested list of length `batch_size,` with each sub-list containing
            denoised
     PIL image sequences of length `num_frames.` It can also be a NumPy array or Torch tensor of shape
    `(batch_size, num_frames, channels, height, width)`
    """

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