Skip to content

Describe Anything

Describe a chosen region of an image or video

Visit official site

What Describe Anything is

Describe Anything is a model that produces detailed descriptions for localised regions of images and videos. It is developed by NVIDIA with several universities and accepts simple region marks.

What you can do with it

  • Generate a detailed description for a region
  • Mark a region with a point, box or scribble
  • Use automatic mask generation

Getting started

  • Install the package from the repository
  • Choose an image or video and mark a region
  • Run the example script and review the description

Pros & cons

✓ What we like

  • Self contained scripts need no extra dependencies
  • Automatic masks simplify region input

! What to watch out for

  • A region mark must be supplied for detailed output
  • Video input needs an annotation on a frame

FAQ

What marks are accepted?

Points, boxes, scribbles or masks.

Who developed it?

NVIDIA together with several universities.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review