Decode a batch of images. Args: z (`torch.FloatTensor`): Input batch of latent vectors. return_dict (`bool`,

Turn a ResNetBlock into a sequence of tasks and append to the task queue @param queue: the target task queue @param block: ResNetBlock

Get mean and var for group norm

Encode a batch of images into latents. Args: x (`torch.FloatTensor`): Input batch of images. return_dict (`b

Encode text prompts into embeddings.

Merge LoRA deltas into UNet weights.

Restore original UNet weights.

Calculate mean and std for adaptive_instance_normalization. Args: feat (Tensor): 4D tensor. eps (float): A small value added to th

Clone a task queue @param task_queue: the task queue to be cloned @return: the cloned task queue

Custom group norm with fixed mean and var @param input: input tensor @param num_groups: number of groups. by default, num_groups = 32

Get the best tile size for GPU memory

r""" Sets the attention processor to use to compute attention. Parameters: processor (`dict` of `AttentionProcessor` or o

r""" Sets the attention processor to use to compute attention. Parameters: processor (`dict` of `AttentionProcessor` or o

Apply wavelet decomposition to the input tensor. This function only returns the low frequency & the high frequency.

Args: x: (B, H, W, C) window_size (int): window size Returns: windows: (num_windows*B, window_size, window_size, C)

Args: x: (B, H, W, C) window_size (int): window size Returns: windows: (num_windows*B, window_size, window_size, C)

Generate a Gaussian mask for tile contributions.

Get the appropriate data type based on precision.

Initialize VAE with tiled encoding/decoding.

Load checkpoint and initialize LoRA adapters.

Load pretrained weights and initialize LoRA adapters.

Load weights from .npz checkpoints for official Google Brain Flax implementation

Move models to the correct device and precision.

Predict on the entire latent without tiling.

Predict on the latent with tiling.

Precompute and store LoRA deltas for the given adapters.

Process latents with or without tiling.

Adaptive instance normalization. Adjust the reference features to have the similar color and illuminations as those in the degradate features.

Build the sampling part of a task queue @param task_queue: the target task queue @param net: the network @param is_decoder: currently

Build a single task queue for the encoder or decoder @param net: the VAE decoder or encoder network @param is_decoder: currently building

Crop the valid region from the tile @param x: input tile @param input_bbox: original input bounding box @param target_bbox: output bo

r""" Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to compute decoding and en

Encode text prompts into embeddings.

create a function from a single tile without summary

Makes broadcastable attention and causal masks so that future and masked tokens are ignored. Arguments: attention_mask (

Makes broadcastable attention and causal masks so that future and masked tokens are ignored. Arguments: attention_mask (