Skip to content

CustomVideo

Generate videos that keep several subject identities

Visit official site

What CustomVideo is

The framework is described as a way to generate videos under multiple subject guidance while keeping identity features intact.

What you can do with it

  • Generate video from multi-subject text guidance
  • Keep identity features across the video
  • Segment objects from reference images

Model notes

  • An attention control strategy separates subjects in latent space
  • Object masks help the model focus on specific regions
  • A benchmark dataset covers 69 subjects and 57 meaningful pairs

Pros & cons

✓ What we like

  • A benchmark dataset with 69 individual subjects is collected
  • 57 meaningful subject pairs are listed in the dataset

! What to watch out for

  • Subject separation relies on the attention control strategy
  • Object masks come from given reference images

FAQ

What is in the dataset?

The material states a multi-subject text-to-video dataset with 69 individual subjects and 57 meaningful pairs.

How are subjects kept apart?

The material describes an attention control strategy in the latent space of the diffusion model.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review