Make-A-Video

Make-A-Video is an AI system that turns text prompts into videos, applying the latest text-to-image generation technology to create fantastical and realistic videos. The system, trained on images accompanied by text descriptions and unlabeled videos, learns to capture the world and its motion, and users can turn their imagination into reality with a few sentences of text.

Some of the key abilities of Make-A-Video are:

• Text-to-Video Generation: Generate videos from text prompts.
• Variations and Motion: Add motion to a single image or fill in the in-between motion to two images.
• Video Variations: Generate multiple variations of a video based on a single input.

The system is capable of generating a broad range of videos, from surreal to realistic and stylized. Examples include a dog in a superhero costume flying through the air, a teddy bear painting a portrait, or a robot dancing in Times Square. Make-A-Video is designed to produce high-quality results, with a 3x improvement in representation and 3x improvement in quality compared to the previous state-of-the-art model.

To develop AI responsibly, Meta AI is working to minimize the generation of harmful or biased content by analyzing millions of data points and applying filters. It also adds watermarks to the generated videos so they can be easily identified as AI-generated content. The technology isn't available to the public today, but the company hopes to release it in the future while ensuring safety and intent.

For people who want to turn text prompts into creative videos, Make-A-Video seems like a good option. The company hasn't yet revealed availability or pricing details. You can check the project's website for details and updates on its development and release.

Published on June 9, 2024