使用 attention 融合 refernet 中的 emb 到 unet 对应的 latens 中 # TODO: 目前只支持 bt hw c 的融合，后续考虑增加对 视频 bhw t c、b thw c的融合 residual_connection: bool = True,

A 2D Transformer model for image-like data. Parameters: num_attention_heads (`int`, *optional*, defaults to 16): The number of heads

wraper of diffusers pipeline, support generation function interface. support 1. text2video: inputs include text, image(optional), refer_image(opti

`DDIMScheduler` extends the denoising procedure introduced in denoising diffusion probabilistic models (DDPMs) with non-Markovian guidance.

r""" Modified Attention class which has special layer, like ip_apadapter_to_k, ip_apadapter_to_v,

属性文本转换功能类，将value作为文本. class for converting attributes to text which only uses the value as text.

`DDPMScheduler` explores the connections between denoising score matching and Langevin dynamics sampling. This model inherits from [`Schedul

Output class for the scheduler's step function output. Args: prev_sample (`torch.FloatTensor` of shape `(batch_size, num_channels, h

将多属性组成的字典转换成完整的文本描述，目前采用简单的前后拼接方式，以`, `作为拼接符号 class for converting a dictionary of multiple attributes into a complete text description. Currently

可以将任务字典转化为形象提词模板类 template class for converting task dictionaries into image prompt templates Args: MultiAttr2PromptTemplate (_typ

The output of [`UNet3DConditionModel`]. Args: sample (`torch.FloatTensor` of shape `(batch_size, num_frames, num_channels, height, w

A 2D UNet mid-block [`UNetMidBlock2D`] with multiple residual blocks and optional attention blocks. Args: in_channels (`int`): The n

属性文本转换功能类，将key和value拼接在一起作为文本. class for converting attributes to text which concatenates the key and value together as text.

属性转化为文本的基类，该类作用就是输入属性，转化为描述文本。 Base class for converting attributes to text which converts attributes to prompt text.

DPM-Solver (and the improved version DPM-Solver++) is a fast dedicated high-order solver for diffusion ODEs with the convergence order guaran

Ancestral sampling with Euler method steps. Based on the original k-diffusion implementation by Katherine Crowson: https://github.com/crowson

将多属性转化为模型输入文本的实际类 The actual class that converts multiple attributes into model input text is

a union diffusers pipeline, support 1. text2image model only, or text2video model, by setting skip_temporal_layer 2. text2video, image2vi

r""" 面向首帧的 referenceonly attn,适用于 T2I的 self_attn referenceonly with vis_cond as key, value, in t2i self_attn.

预置了多种关注属性转换的类，方便维护 class for multiple attribute conversion with multiple attention attributes preset for easy maintenance

继承 UNet2DConditionModel. 新增功能，类似controlnet 返回模型中间特征，用于后续作用 Inherit Unet2DConditionModel. Add new functions, similar to controlnet, return the

继承 UNet3DConditionModel， 用于提取中间emb用于后续作用。 Inherit Unet3DConditionModel, used to extract the middle emb for subsequent actions. Args:

封装了各种子模型的超模型，与 diffusers 的 pipeline 很像，只不过这里是模型定义。 主要作用 1. 将支持controlnet、referencenet等功能的计算封装起来，简洁些； 2. 便于 accelerator 的分布式训练； wrap t

r""" 面向 ref_image的 self_attn的 IPAdapter

Temporal convolutional layer that can be used for video (sequence of images) input Code mostly copied from: https://github.com/modelscope/mod

Transformer model for video-like data. Parameters: num_attention_heads (`int`, *optional*, defaults to 16): The number of heads to u

r""" UNet3DConditionModel is a conditional 2D UNet model that takes in a noisy sample, conditional state, and a timestep and returns sample sh