Skip to content

MarDini

Generate video from masked frames with diffusion

Visit official site

What MarDini is

The model is described as a video diffusion model that combines masked autoregression with a unified diffusion framework for masked-frame video generation.

What you can do with it

  • Fill middle frames between a first and last frame
  • Generate video starting from a single image
  • Extend an existing video with new masked frames
  • Build longer clips through recursive interpolation

Getting started

  • Visit the product page and download the models and code
  • Set the task with a masking strategy and video parameters
  • Prepare an image or video sequence in the required format
  • Run inference to produce the video content

Pros & cons

✓ What we like

  • Most compute goes to a low-resolution planning model
  • Recursive interpolation builds longer clips from few images

! What to watch out for

  • Tasks depend on the chosen masking strategy
  • 3D view synthesis is described as an early capability

FAQ

Which tasks are listed?

The feature list names interpolation, image-to-video generation and video extension.

Who released it?

The material states it comes from Meta AI Research.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review