Fast autoregressive generation with CUDA-graphed predictor and talker.

Converts all `BatchNorm2d` and `SyncBatchNorm` layers of provided module into `FrozenBatchNorm2d`. If `module` is itself an instance of eithe

Generate speech with voice cloning using reference audio. Args: text: Text to synthesize language: Target la

Compute the intermediate time steps for sampling.

image[0, 1] -> latent

Args: text_tokens: a 1D (unbatched) or 2D (batched) tensor.

lock modules Args: unlocked_groups (int): leave last n layer groups unlocked (default: 0)

Performs log-linear interpolation of a given array of decreasing numbers.

Compute alpha_t of a given continuous-time label t in [0, T].

Convert bool-tensor to float-tensor for flash attention. Parameters ---------- lengths (torch.Tensor): Batch of lengths (B, ?).

Merges tokenized outputs by keeping the middle and dropping half of the overlapped tokens. Args: - tokenized_segments (List[List[int]]):

Parse ControlNet infotext string and write result to `results` dict.

Called in inpainting mode when the original content is blended with the inpainted content. This is called at every step in the deno

parses generation parameters string, the one you see in text field under the picture in UI: ``` girl with an artist's beret, determined, blue eyes,

converts a line from config ("ext1 ext2, ext3 ") into a python list (["ext1", "ext2", "ext3"])

Prepares an embedded sequence tensor by applying positional embeddings and handling padding masks. Args: x_B_C_T_H_W

This function is called before processing begins for AlwaysVisible scripts. You can modify the processing object (p) here, inject h

registers extra networks page for the UI; recommend doing it in on_before_ui() callback for extensions

Clears the style database and reloads the styles from the CSV file(s) matching the path used to initialize the database.