import copy
import inspect
from dataclasses import dataclass
from typing import Callable, List, Optional, Union

import numpy as np
import PIL.Image
import torch
import torch.nn.functional as F
from torch.nn.functional import grid_sample
from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer

from ...image_processor import VaeImageProcessor
from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
from ...models import AutoencoderKL, UNet2DConditionModel
from ...models.lora import adjust_lora_scale_text_encoder
from ...schedulers import KarrasDiffusionSchedulers
from ...utils import USE_PEFT_BACKEND, BaseOutput, logging, scale_lora_layers, unscale_lora_layers
from ...utils.torch_utils import randn_tensor
from ..pipeline_utils import DiffusionPipeline
from ..stable_diffusion import StableDiffusionSafetyChecker


logger = logging.get_logger(__name__)  # pylint: disable=invalid-name


def rearrange_0(tensor, f):
    F, C, H, W = tensor.size()
    tensor = torch.permute(torch.reshape(tensor, (F // f, f, C, H, W)), (0, 2, 1, 3, 4))
    return tensor


def rearrange_1(tensor):
    B, C, F, H, W = tensor.size()
    return torch.reshape(torch.permute(tensor, (0, 2, 1, 3, 4)), (B * F, C, H, W))


def rearrange_3(tensor, f):
    F, D, C = tensor.size()
    return torch.reshape(tensor, (F // f, f, D, C))


def rearrange_4(tensor):
    B, F, D, C = tensor.size()
    return torch.reshape(tensor, (B * F, D, C))


class CrossFrameAttnProcessor:
    """
    Cross frame attention processor. Each frame attends the first frame.

    Args:
        batch_size: The number that represents actual batch size, other than the frames.
            For example, calling unet with a single prompt and num_images_per_prompt=1, batch_size should be equal to
            2, due to classifier-free guidance.
    """

    def __init__(self, batch_size=2):
        self.batch_size = batch_size

    def __call__(self, attn, hidden_states, encoder_hidden_states=None, attention_mask=None):
        batch_size, sequence_length, _ = hidden_states.shape
        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
        query = attn.to_q(hidden_states)

        is_cross_attention = encoder_hidden_states is not None
        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        # Cross Frame Attention
        if not is_cross_attention:
            video_length = key.size()[0] // self.batch_size
            first_frame_index = [0] * video_length

            # rearrange keys to have batch and frames in the 1st and 2nd dims respectively
            key = rearrange_3(key, video_length)
            key = key[:, first_frame_index]
            # rearrange values to have batch and frames in the 1st and 2nd dims respectively
            value = rearrange_3(value, video_length)
            value = value[:, first_frame_index]

            # rearrange back to original shape
            key = rearrange_4(key)
            value = rearrange_4(value)

        query = attn.head_to_batch_dim(query)
        key = attn.head_to_batch_dim(key)
        value = attn.head_to_batch_dim(value)

        attention_probs = attn.get_attention_scores(query, key, attention_mask)
        hidden_states = torch.bmm(attention_probs, value)
        hidden_states = attn.batch_to_head_dim(hidden_states)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)

        return hidden_states


class CrossFrameAttnProcessor2_0:
    """
    Cross frame attention processor with scaled_dot_product attention of Pytorch 2.0.

    Args:
        batch_size: The number that represents actual batch size, other than the frames.
            For example, calling unet with a single prompt and num_images_per_prompt=1, batch_size should be equal to
            2, due to classifier-free guidance.
    """

    def __init__(self, batch_size=2):
        if not hasattr(F, "scaled_dot_product_attention"):
            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
        self.batch_size = batch_size

    def __call__(self, attn, hidden_states, encoder_hidden_states=None, attention_mask=None):
        batch_size, sequence_length, _ = (
            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
        )
        inner_dim = hidden_states.shape[-1]

        if attention_mask is not None:
            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
            # scaled_dot_product_attention expects attention_mask shape to be
            # (batch, heads, source_length, target_length)
            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])

        query = attn.to_q(hidden_states)

        is_cross_attention = encoder_hidden_states is not None
        if encoder_hidden_states is None:
            encoder_hidden_states = hidden_states
        elif attn.norm_cross:
            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)

        key = attn.to_k(encoder_hidden_states)
        value = attn.to_v(encoder_hidden_states)

        # Cross Frame Attention
        if not is_cross_attention:
            video_length = max(1, key.size()[0] // self.batch_size)
            first_frame_index = [0] * video_length

            # rearrange keys to have batch and frames in the 1st and 2nd dims respectively
            key = rearrange_3(key, video_length)
            key = key[:, first_frame_index]
            # rearrange values to have batch and frames in the 1st and 2nd dims respectively
            value = rearrange_3(value, video_length)
            value = value[:, first_frame_index]

            # rearrange back to original shape
            key = rearrange_4(key)
            value = rearrange_4(value)

        head_dim = inner_dim // attn.heads
        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)

        # the output of sdp = (batch, num_heads, seq_len, head_dim)
        # TODO: add support for attn.scale when we move to Torch 2.1
        hidden_states = F.scaled_dot_product_attention(
            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
        )

        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
        hidden_states = hidden_states.to(query.dtype)

        # linear proj
        hidden_states = attn.to_out[0](hidden_states)
        # dropout
        hidden_states = attn.to_out[1](hidden_states)
        return hidden_states


@dataclass
class TextToVideoPipelineOutput(BaseOutput):
    r"""
    Output class for zero-shot text-to-video pipeline.

    Args:
        images (`[List[PIL.Image.Image]`, `np.ndarray`]):
            List of denoised PIL images of length `batch_size` or NumPy array of shape `(batch_size, height, width,
            num_channels)`.
        nsfw_content_detected (`[List[bool]]`):
            List indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content or
            `None` if safety checking could not be performed.
    """

    images: Union[List[PIL.Image.Image], np.ndarray]
    nsfw_content_detected: Optional[List[bool]]


def coords_grid(batch, ht, wd, device):
    # Adapted from https://github.com/princeton-vl/RAFT/blob/master/core/utils/utils.py
    coords = torch.meshgrid(torch.arange(ht, device=device), torch.arange(wd, device=device))
    coords = torch.stack(coords[::-1], dim=0).float()
    return coords[None].repeat(batch, 1, 1, 1)


def warp_single_latent(latent, reference_flow):
    """
    Warp latent of a single frame with given flow

    Args:
        latent: latent code of a single frame
        reference_flow: flow which to warp the latent with

    Returns:
        warped: warped latent
    """
    _, _, H, W = reference_flow.size()
    _, _, h, w = latent.size()
    coords0 = coords_grid(1, H, W, device=latent.device).to(latent.dtype)

    coords_t0 = coords0 + reference_flow
    coords_t0[:, 0] /= W
    coords_t0[:, 1] /= H

    coords_t0 = coords_t0 * 2.0 - 1.0
    coords_t0 = F.interpolate(coords_t0, size=(h, w), mode="bilinear")
    coords_t0 = torch.permute(coords_t0, (0, 2, 3, 1))

    warped = grid_sample(latent, coords_t0, mode="nearest", padding_mode="reflection")
    return warped


def create_motion_field(motion_field_strength_x, motion_field_strength_y, frame_ids, device, dtype):
    """
    Create translation motion field

    Args:
        motion_field_strength_x: motion strength along x-axis
        motion_field_strength_y: motion strength along y-axis
        frame_ids: indexes of the frames the latents of which are being processed.
            This is needed when we perform chunk-by-chunk inference
        device: device
        dtype: dtype

    Returns:

    """
    seq_length = len(frame_ids)
    reference_flow = torch.zeros((seq_length, 2, 512, 512), device=device, dtype=dtype)
    for fr_idx in range(seq_length):
        reference_flow[fr_idx, 0, :, :] = motion_field_strength_x * (frame_ids[fr_idx])
        reference_flow[fr_idx, 1, :, :] = motion_field_strength_y * (frame_ids[fr_idx])
    return reference_flow


def create_motion_field_and_warp_latents(motion_field_strength_x, motion_field_strength_y, frame_ids, latents):
    """
    Creates translation motion and warps the latents accordingly

    Args:
        motion_field_strength_x: motion strength along x-axis
        motion_field_strength_y: motion strength along y-axis
        frame_ids: indexes of the frames the latents of which are being processed.
            This is needed when we perform chunk-by-chunk inference
        latents: latent codes of frames

    Returns:
        warped_latents: warped latents
    """
    motion_field = create_motion_field(
        motion_field_strength_x=motion_field_strength_x,
        motion_field_strength_y=motion_field_strength_y,
        frame_ids=frame_ids,
        device=latents.device,
        dtype=latents.dtype,
    )
    warped_latents = latents.clone().detach()
    for i in range(len(warped_latents)):
        warped_latents[i] = warp_single_latent(latents[i][None], motion_field[i][None])
    return warped_latents


class TextToVideoZeroPipeline(DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin):
    r"""
    Pipeline for zero-shot text-to-video generation using Stable 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 ([`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 ([`UNet2DConditionModel`]):
            A [`UNet3DConditionModel`] to denoise the encoded video latents.
        scheduler ([`SchedulerMixin`]):
            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
        safety_checker ([`StableDiffusionSafetyChecker`]):
            Classification module that estimates whether generated images could be considered offensive or harmful.
            Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details
            about a model's potential harms.
        feature_extractor ([`CLIPImageProcessor`]):
            A [`CLIPImageProcessor`] to extract features from generated images; used as inputs to the `safety_checker`.
    """

    def __init__(
        self,
        vae: AutoencoderKL,
        text_encoder: CLIPTextModel,
        tokenizer: CLIPTokenizer,
        unet: UNet2DConditionModel,
        scheduler: KarrasDiffusionSchedulers,
        safety_checker: StableDiffusionSafetyChecker,
        feature_extractor: CLIPImageProcessor,
        requires_safety_checker: bool = True,
    ):
        super().__init__()
        self.register_modules(
            vae=vae,
            text_encoder=text_encoder,
            tokenizer=tokenizer,
            unet=unet,
            scheduler=scheduler,
            safety_checker=safety_checker,
            feature_extractor=feature_extractor,
        )
        processor = (
            CrossFrameAttnProcessor2_0(batch_size=2)
            if hasattr(F, "scaled_dot_product_attention")
            else CrossFrameAttnProcessor(batch_size=2)
        )
        self.unet.set_attn_processor(processor)

        if safety_checker is None and requires_safety_checker:
            logger.warning(
                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
            )
        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)

    def forward_loop(self, x_t0, t0, t1, generator):
        """
        Perform DDPM forward process from time t0 to t1. This is the same as adding noise with corresponding variance.

        Args:
            x_t0:
                Latent code at time t0.
            t0:
                Timestep at t0.
            t1:
                Timestamp at t1.
            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                generation deterministic.

        Returns:
            x_t1:
                Forward process applied to x_t0 from time t0 to t1.
        """
        eps = randn_tensor(x_t0.size(), generator=generator, dtype=x_t0.dtype, device=x_t0.device)
        alpha_vec = torch.prod(self.scheduler.alphas[t0:t1])
        x_t1 = torch.sqrt(alpha_vec) * x_t0 + torch.sqrt(1 - alpha_vec) * eps
        return x_t1

    def backward_loop(
        self,
        latents,
        timesteps,
        prompt_embeds,
        guidance_scale,
        callback,
        callback_steps,
        num_warmup_steps,
        extra_step_kwargs,
        cross_attention_kwargs=None,
    ):
        """
        Perform backward process given list of time steps.

        Args:
            latents:
                Latents at time timesteps[0].
            timesteps:
                Time steps along which to perform backward process.
            prompt_embeds:
                Pre-generated text embeddings.
            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`.
            callback (`Callable`, *optional*):
                A function that calls every `callback_steps` steps during inference. The function is called with the
                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
            callback_steps (`int`, *optional*, defaults to 1):
                The frequency at which the `callback` function is called. If not specified, the callback is called at
                every step.
            extra_step_kwargs:
                Extra_step_kwargs.
            cross_attention_kwargs:
                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).
            num_warmup_steps:
                number of warmup steps.

        Returns:
            latents:
                Latents of backward process output at time timesteps[-1].
        """
        do_classifier_free_guidance = guidance_scale > 1.0
        num_steps = (len(timesteps) - num_warmup_steps) // self.scheduler.order
        with self.progress_bar(total=num_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                # expand the latents if we are doing classifier free guidance
                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)

                # predict the noise residual
                noise_pred = self.unet(
                    latent_model_input,
                    t,
                    encoder_hidden_states=prompt_embeds,
                    cross_attention_kwargs=cross_attention_kwargs,
                ).sample

                # perform guidance
                if 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)

                # compute the previous noisy sample x_t -> x_t-1
                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample

                # 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()
                    if callback is not None and i % callback_steps == 0:
                        step_idx = i // getattr(self.scheduler, "order", 1)
                        callback(step_idx, t, latents)
        return latents.clone().detach()

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.check_inputs
    def check_inputs(
        self,
        prompt,
        height,
        width,
        callback_steps,
        negative_prompt=None,
        prompt_embeds=None,
        negative_prompt_embeds=None,
        callback_on_step_end_tensor_inputs=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 callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
            raise ValueError(
                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
                f" {type(callback_steps)}."
            )
        if callback_on_step_end_tensor_inputs is not None and not all(
            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
        ):
            raise ValueError(
                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
            )

        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}."
                )

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
        shape = (batch_size, num_channels_latents, 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, device=device, dtype=dtype)
        else:
            latents = latents.to(device)

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

    @torch.no_grad()
    def __call__(
        self,
        prompt: Union[str, List[str]],
        video_length: Optional[int] = 8,
        height: Optional[int] = None,
        width: Optional[int] = None,
        num_inference_steps: int = 50,
        guidance_scale: float = 7.5,
        negative_prompt: Optional[Union[str, List[str]]] = None,
        num_videos_per_prompt: Optional[int] = 1,
        eta: float = 0.0,
        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
        latents: Optional[torch.FloatTensor] = None,
        motion_field_strength_x: float = 12,
        motion_field_strength_y: float = 12,
        output_type: Optional[str] = "tensor",
        return_dict: bool = True,
        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
        callback_steps: Optional[int] = 1,
        t0: int = 44,
        t1: int = 47,
        frame_ids: Optional[List[int]] = None,
    ):
        """
        The call function to the pipeline for generation.

        Args:
            prompt (`str` or `List[str]`, *optional*):
                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
            video_length (`int`, *optional*, defaults to 8):
                The number of generated video frames.
            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_inference_steps (`int`, *optional*, defaults to 50):
                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                expense of slower inference.
            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 video generation. If not defined, you need to
                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
            num_videos_per_prompt (`int`, *optional*, defaults to 1):
                The number of videos to generate per prompt.
            eta (`float`, *optional*, defaults to 0.0):
                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.
            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
                generation deterministic.
            latents (`torch.FloatTensor`, *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 `"numpy"`):
                The output format of the generated video. Choose between `"latent"` and `"numpy"`.
            return_dict (`bool`, *optional*, defaults to `True`):
                Whether or not to return a
                [`~pipelines.text_to_video_synthesis.pipeline_text_to_video_zero.TextToVideoPipelineOutput`] instead of
                a plain tuple.
            callback (`Callable`, *optional*):
                A function that calls every `callback_steps` steps during inference. The function is called with the
                following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
            callback_steps (`int`, *optional*, defaults to 1):
                The frequency at which the `callback` function is called. If not specified, the callback is called at
                every step.
            motion_field_strength_x (`float`, *optional*, defaults to 12):
                Strength of motion in generated video along x-axis. See the [paper](https://arxiv.org/abs/2303.13439),
                Sect. 3.3.1.
            motion_field_strength_y (`float`, *optional*, defaults to 12):
                Strength of motion in generated video along y-axis. See the [paper](https://arxiv.org/abs/2303.13439),
                Sect. 3.3.1.
            t0 (`int`, *optional*, defaults to 44):
                Timestep t0. Should be in the range [0, num_inference_steps - 1]. See the
                [paper](https://arxiv.org/abs/2303.13439), Sect. 3.3.1.
            t1 (`int`, *optional*, defaults to 47):
                Timestep t0. Should be in the range [t0 + 1, num_inference_steps - 1]. See the
                [paper](https://arxiv.org/abs/2303.13439), Sect. 3.3.1.
            frame_ids (`List[int]`, *optional*):
                Indexes of the frames that are being generated. This is used when generating longer videos
                chunk-by-chunk.

        Returns:
            [`~pipelines.text_to_video_synthesis.pipeline_text_to_video_zero.TextToVideoPipelineOutput`]:
                The output contains a `ndarray` of the generated video, when `output_type` != `"latent"`, otherwise a
                latent code of generated videos and a list of `bool`s indicating whether the corresponding generated
                video contains "not-safe-for-work" (nsfw) content..
        """
        assert video_length > 0
        if frame_ids is None:
            frame_ids = list(range(video_length))
        assert len(frame_ids) == video_length

        assert num_videos_per_prompt == 1

        if isinstance(prompt, str):
            prompt = [prompt]
        if isinstance(negative_prompt, str):
            negative_prompt = [negative_prompt]

        # 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

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

        # Define call parameters
        batch_size = 1 if isinstance(prompt, str) else len(prompt)
        device = self._execution_device
        # 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.
        do_classifier_free_guidance = guidance_scale > 1.0

        # Encode input prompt
        prompt_embeds_tuple = self.encode_prompt(
            prompt, device, num_videos_per_prompt, do_classifier_free_guidance, negative_prompt
        )
        prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]])

        # Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, device=device)
        timesteps = self.scheduler.timesteps

        # Prepare latent variables
        num_channels_latents = self.unet.config.in_channels
        latents = self.prepare_latents(
            batch_size * num_videos_per_prompt,
            num_channels_latents,
            height,
            width,
            prompt_embeds.dtype,
            device,
            generator,
            latents,
        )
        # Prepare extra step kwargs.
        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order

        # Perform the first backward process up to time T_1
        x_1_t1 = self.backward_loop(
            timesteps=timesteps[: -t1 - 1],
            prompt_embeds=prompt_embeds,
            latents=latents,
            guidance_scale=guidance_scale,
            callback=callback,
            callback_steps=callback_steps,
            extra_step_kwargs=extra_step_kwargs,
            num_warmup_steps=num_warmup_steps,
        )
        scheduler_copy = copy.deepcopy(self.scheduler)

        # Perform the second backward process up to time T_0
        x_1_t0 = self.backward_loop(
            timesteps=timesteps[-t1 - 1 : -t0 - 1],
            prompt_embeds=prompt_embeds,
            latents=x_1_t1,
            guidance_scale=guidance_scale,
            callback=callback,
            callback_steps=callback_steps,
            extra_step_kwargs=extra_step_kwargs,
            num_warmup_steps=0,
        )

        # Propagate first frame latents at time T_0 to remaining frames
        x_2k_t0 = x_1_t0.repeat(video_length - 1, 1, 1, 1)

        # Add motion in latents at time T_0
        x_2k_t0 = create_motion_field_and_warp_latents(
            motion_field_strength_x=motion_field_strength_x,
            motion_field_strength_y=motion_field_strength_y,
            latents=x_2k_t0,
            frame_ids=frame_ids[1:],
        )

        # Perform forward process up to time T_1
        x_2k_t1 = self.forward_loop(
            x_t0=x_2k_t0,
            t0=timesteps[-t0 - 1].item(),
            t1=timesteps[-t1 - 1].item(),
            generator=generator,
        )

        # Perform backward process from time T_1 to 0
        x_1k_t1 = torch.cat([x_1_t1, x_2k_t1])
        b, l, d = prompt_embeds.size()
        prompt_embeds = prompt_embeds[:, None].repeat(1, video_length, 1, 1).reshape(b * video_length, l, d)

        self.scheduler = scheduler_copy
        x_1k_0 = self.backward_loop(
            timesteps=timesteps[-t1 - 1 :],
            prompt_embeds=prompt_embeds,
            latents=x_1k_t1,
            guidance_scale=guidance_scale,
            callback=callback,
            callback_steps=callback_steps,
            extra_step_kwargs=extra_step_kwargs,
            num_warmup_steps=0,
        )
        latents = x_1k_0

        # manually for max memory savings
        if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
            self.unet.to("cpu")
        torch.cuda.empty_cache()

        if output_type == "latent":
            image = latents
            has_nsfw_concept = None
        else:
            image = self.decode_latents(latents)
            # Run safety checker
            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)

        # Offload all models
        self.maybe_free_model_hooks()

        if not return_dict:
            return (image, has_nsfw_concept)

        return TextToVideoPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
    def run_safety_checker(self, image, device, dtype):
        if self.safety_checker is None:
            has_nsfw_concept = None
        else:
            if torch.is_tensor(image):
                feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
            else:
                feature_extractor_input = self.image_processor.numpy_to_pil(image)
            safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
            image, has_nsfw_concept = self.safety_checker(
                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
            )
        return image, has_nsfw_concept

    # 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

    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt
    def encode_prompt(
        self,
        prompt,
        device,
        num_images_per_prompt,
        do_classifier_free_guidance,
        negative_prompt=None,
        prompt_embeds: Optional[torch.FloatTensor] = None,
        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
        lora_scale: Optional[float] = 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
            device: (`torch.device`):
                torch device
            num_images_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 (`torch.FloatTensor`, *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 (`torch.FloatTensor`, *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.
            lora_scale (`float`, *optional*):
                A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
            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.
        """
        # set lora scale so that monkey patched LoRA
        # function of text encoder can correctly access it
        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
            self._lora_scale = lora_scale

            # dynamically adjust the LoRA scale
            if not USE_PEFT_BACKEND:
                adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
            else:
                scale_lora_layers(self.text_encoder, lora_scale)

        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:
            # textual inversion: procecss multi-vector tokens if necessary
            if isinstance(self, TextualInversionLoaderMixin):
                prompt = self.maybe_convert_prompt(prompt, self.tokenizer)

            text_inputs = self.tokenizer(
                prompt,
                padding="max_length",
                max_length=self.tokenizer.model_max_length,
                truncation=True,
                return_tensors="pt",
            )
            text_input_ids = text_inputs.input_ids
            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids

            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.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.to(device)
            else:
                attention_mask = None

            if clip_skip is None:
                prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask)
                prompt_embeds = prompt_embeds[0]
            else:
                prompt_embeds = self.text_encoder(
                    text_input_ids.to(device), 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, device=device)

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

        # get unconditional embeddings for classifier free guidance
        if 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

            # textual inversion: procecss multi-vector tokens if necessary
            if isinstance(self, TextualInversionLoaderMixin):
                uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)

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

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

            negative_prompt_embeds = self.text_encoder(
                uncond_input.input_ids.to(device),
                attention_mask=attention_mask,
            )
            negative_prompt_embeds = negative_prompt_embeds[0]

        if 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, device=device)

            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)

        if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
            # Retrieve the original scale by scaling back the LoRA layers
            unscale_lora_layers(self.text_encoder, lora_scale)

        return prompt_embeds, negative_prompt_embeds

    def decode_latents(self, latents):
        latents = 1 / self.vae.config.scaling_factor * latents
        image = self.vae.decode(latents, return_dict=False)[0]
        image = (image / 2 + 0.5).clamp(0, 1)
        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
        return image