fastvideo.v1.pipelines.stages.latent_preparation

Latent preparation stage for diffusion pipelines.

Module Contents

Classes

LatentPreparationStage

Stage for preparing initial latent variables for the diffusion process.

Data

logger

API

class fastvideo.v1.pipelines.stages.latent_preparation.LatentPreparationStage(scheduler, transformer)

Bases: fastvideo.v1.pipelines.stages.base.PipelineStage

Stage for preparing initial latent variables for the diffusion process.

This stage handles the preparation of the initial latent variables that will be denoised during the diffusion process.

Initialization

adjust_video_length(batch: fastvideo.v1.pipelines.pipeline_batch_info.ForwardBatch, fastvideo_args: fastvideo.v1.fastvideo_args.FastVideoArgs) → int

Adjust video length based on VAE version.

Parameters:

• batch – The current batch information.

• fastvideo_args – The inference arguments.

Returns:

The batch with adjusted video length.

forward(batch: fastvideo.v1.pipelines.pipeline_batch_info.ForwardBatch, fastvideo_args: fastvideo.v1.fastvideo_args.FastVideoArgs) → fastvideo.v1.pipelines.pipeline_batch_info.ForwardBatch

Prepare initial latent variables for the diffusion process.

Parameters:

• batch – The current batch information.

• fastvideo_args – The inference arguments.

Returns:

The batch with prepared latent variables.

fastvideo.v1.pipelines.stages.latent_preparation.logger

‘init_logger(…)’