Skip to content

Segment Anything

Segment any object in an image with a prompt

Visit official site

What Segment Anything is

The model is described as a promptable segmentation system that generalises to unfamiliar objects and images without extra training. The site also describes one-click object cutout from an uploaded image.

What you can do with it

  • Use interactive points and boxes as prompts
  • Segment everything in an image automatically
  • Generate several valid masks for an ambiguous prompt
  • Segment an object by text instruction after clicking it

Application scenarios

  • Select a dog in an image for segmentation
  • Use an object detection box for text-to-object segmentation
  • Edit images with segmentation masks
  • Apply the model to image segmentation, image editing and video object tracking

Model notes

  • Zero-shot generalisation needs no extra training
  • Training is stated as 11 million images and 1 billion masks
  • A modular design allows inference within a few milliseconds
  • The promptable design can be integrated with other systems

Pros & cons

✓ What we like

  • Prompts include interactive points, boxes and text instructions
  • The promptable design can be integrated with other systems

! What to watch out for

  • An ambiguous prompt can return several valid masks
  • The described cutout flow starts with an uploaded image and a clicked object

FAQ

What is SAM?

The material describes a promptable segmentation system with zero-shot generalisation.

What data was it trained on?

The material states 11 million images and 1 billion segmentation masks.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review