Skip to content

Tora

Control video motion with trajectory guided diffusion transformers

Visit official site

What Tora is

Tora is a video generation model built on diffusion transformers, or DiT, that integrates text, visual and trajectory conditions for precise motion control. The model is described as using the scalability of DiT for video across different durations, aspect ratios and resolutions.

What you can do with it

  • Encode any trajectory into hierarchical spatial temporal motion patches
  • Keep motion information between consecutive frames with a spatial-temporal DiT
  • Inject multi-level motion conditions into DiT blocks with a motion guidance fuser
  • Generate video at different resolutions and extended durations

Example uses

  • A landscape video of drifting roses against a snowy mountain background
  • An underwater scene with a flying seagull and colourful coral reefs
  • A commercial clip with a red helium balloon rising over a desert

Getting started

  • Set the trajectory and motion requirements for the video
  • Encode the trajectory into spatial temporal motion patches with TE
  • Generate an initial video sketch with the spatial-temporal diffusion transformer
  • Inject motion conditions into DiT blocks with MGF

Pros & cons

✓ What we like

  • Text, visual and trajectory conditions are combined
  • Different durations, aspect ratios and resolutions are supported

! What to watch out for

  • Motion control follows the trajectory conditions supplied
  • The pipeline runs through TE, DiT and MGF stages

FAQ

What conditions does it use?

The material names text, visual and trajectory conditions integrated for motion control.

What are the main components?

A trajectory extractor, a spatial-temporal diffusion transformer with a 3D video compression network, and a motion guidance fuser.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review