Skip to content

ActAnywhere

Generate video backgrounds that follow a foreground subject

Visit official site

What ActAnywhere is

The model is described as a subject-aware generative model for video backgrounds, built by tailoring a large-scale video diffusion model to the task. The model takes foreground subject segmentations with a conditioning image and returns a coherent video aligned with the condition frame.

What you can do with it

  • Generate a video background from a foreground segmentation sequence and a condition image
  • Coordinate the background with foreground motion and appearance
  • Use a composite frame with the subject or a background-only frame as the condition image
  • Produce video backgrounds with different camera motions

Example uses

  • Combine a moving-person segmentation with a seaside image to show a person running by the sea
  • Combine a dance segmentation with an ancient palace image to stage the dance inside a palace
  • Combine a driving-car segmentation with a skyline image to show a car in city streets

Model notes

  • A large-scale video diffusion model is customised for background generation
  • The model is trained on a large-scale human-object interaction video dataset
  • The material reports evaluations in which the model clearly outperforms a baseline
  • The material states generalisation to varied samples including non-human subjects

Pros & cons

✓ What we like

  • A large-scale video diffusion model is tailored to the task
  • Realistic foreground and background interaction is described in the output

! What to watch out for

  • Input requires a sequence of foreground subject segmentations
  • Generation is conditioned on an image describing the wanted scene

FAQ

What does the model take as input?

A sequence of foreground subject segmentations and an image describing the wanted scene.

What is it trained on?

The material states a large-scale human-object interaction video dataset.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review