Precompute the frequency tensor for complex exponentials (cis) with given dimensions. This function calculates a frequency tensor with compl

Helper function to get activation function from string. Args: act_fn (str): Name of activation function. Returns: nn.Module:

Get current device id based on device type.

Get the normalization layer. Args: norm_layer (str): The type of normalization layer. Returns: norm_layer (nn.Module):

Map latent-space `[start, end)` coordinates to their pixel-space equivalents by scaling each axis (frame/time, height, width) with the corres

r""" Get the attention processor in use. Args: return_deprecated_lora (`bool`, *optional*, defaults to `False`):

Copied from https://github.com/hehao13/CameraCtrl/blob/main/inference.py

This matches the implementation in Denoising Diffusion Probabilistic Models: Create sinusoidal timestep embeddings. Args timesteps (t

This matches the implementation in Denoising Diffusion Probabilistic Models: Create sinusoidal timestep embeddings. Args timesteps (

Insert WanVideoUnit_GBufferEncoder into the pipeline's unit list, after ImageEmbedderFused and before FunControl.

features: shape=[1, T, 512] input_fps: fps for audio, f_a output_fps: fps for video, f_m output_len: video length

Look up the prompt for a given clip_dir from metadata_with_captions.csv.

Merge the video and audio into a new video, with the duration set to the shorter of the two, and overwrite the original video file. Para

r""" Normalize the encoder hidden states. Requires `self.norm_cross` to be specified when constructing the `Attention` class.