Skip to content

MotionDirector

Customise the motion of a text-to-video diffusion model

Visit official site

What MotionDirector is

The technique is described as customising text-to-video diffusion models so they generate video with a chosen motion.

What you can do with it

  • Learn the look and motion of a reference video
  • Mix appearance and motion across videos
  • Add a custom motion to a single image

How training and inference split

  • Spatial modules adapt the reference appearance
  • Temporal modules learn the motion dynamics
  • Trained temporal modules are injected at inference

Pros & cons

✓ What we like

  • Appearance and motion are learned separately
  • Trained temporal modules transfer motion to new looks

! What to watch out for

  • Custom motion needs a training stage
  • The method works on text-to-video diffusion models

FAQ

What is the core design?

A dual-path adapter setup that separates appearance from motion.

What are the named applications?

Mixing appearance and motion across videos and animating a single image with a custom motion.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review