Make-A-Video

Make-A-Video is a research AI system from Meta designed to generate short video clips directly from text prompts or still images. Built on advances in text-to-image diffusion models combined with temporal learning, it allows users to describe a scene in natural language and receive an AI-generated video that matches the prompt. The system can create original videos from scratch, extend static images into moving scenes, and apply motion to existing content, demonstrating an understanding of objects, environments, and basic physical dynamics.

Make-A-Video was released as a research demo, aimed primarily at AI researchers, creative technologists, and media innovation teams exploring generative video. The website showcases sample outputs, explains the underlying architecture at a high level, and provides details about the training approach, including the use of unlabeled video data. While not a commercial product with a full-featured interface, it illustrates the potential for future creative tools that rapidly prototype concepts, storyboards, and visual ideas by converting natural language descriptions into video sequences.