Meta Presents Text-to-Video AI Generator

A team of machine learning engineers from Meta has presented a new system dubbed Make-A-Video. It makes short videos that match text descriptions.

Meta's blog reports on the work:

With just a few words or lines of text, Make-A-Video can bring imagination to life and create one-of-a-kind videos full of vivid colors and landscapes.

The videos turn out artificial, with blurred subjects and distorted animation, but still represent a significant achievement in the field of AI content creation.

The videos recorded with the new AI last no more than five seconds and are of low quality, while the neural network, according to the developers, quite accurately reads the description and creates content based on it. It is not yet possible to try the technology out for yourself.

Technically, the system analyses a text query and generates 16 frames of video with a resolution of 64 by 64 pixels each. These are then magnified using another AI system. Allegedly, the neural network has been trained on millions of different videos.