Facebook published an AI model in April 2023 that can pick out individual objects from within an image. It has a dataset of image annotations, which is the largest of its kind. Its SAM or Segment Anything Model could identify objects in images and videos, even for those cases which it has not one across during its training.

A SAM user can select objects clicking on them or by writing text prompts. To illustrate, if the word ‘cat’ is written, the model draws boxes around each of the several cats in a photo.

Facebook’s model spins up surrealist videos from text prompts and generates children’s book illustrations from text prompts.

Facebook was already using a SAM like technology internally to tag photos, moderate prohibited content and to provide recommendations for post on Instagram.

Runway, a NY start up, generated a short video in a couple of minutes of a calm river flowing in a forest on receiving a short description. Runway is testing AI technology that will let people generate videos by typing some words into a box on a computer screen.