Skip to content

AsyncDiff

Asynchronous denoising to parallelise diffusion models

Visit official site

What AsyncDiff is

The project is described as an asynchronous denoising acceleration scheme that splits the noise prediction model across devices to parallelise diffusion inference.

What you can do with it

  • Split noise prediction across devices
  • Reduce inference latency
  • Configure split count and steps

Application scenarios

  • Speed up Stable Diffusion XL images
  • Parallelise ControlNet for video
  • Accelerate the x4 Upscaler

Pros & cons

✓ What we like

  • Cuts latency with little quality loss
  • Flexible parallel configuration

! What to watch out for

  • Requires NVIDIA GPU, CUDA and CuDNN
  • Setup needs split and step configuration

FAQ

Which models are named?

Stable Diffusion XL, ControlNet, AnimateDiff and others.

What hardware is listed?

The steps name an NVIDIA GPU, CUDA and CuDNN.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review