Meta's Muse Video Generates 10-Second Clips

Meta is about to launch a video generation model called Muse Video, and it's already in beta with some partners. The model can make videos up to 10 seconds long in one go. It shows promise in details, understanding scenes and objects, and keeping things stable over time. In tests, it made quality clips like a woman walking down a Tokyo street at night or a rocket exploding into confetti, and it can even mimic the shaky look of phone videos. Meta admits the model still has issues with syncing audio and visuals, and it's not great at fast-moving physics. With only 10-second clips so far, it's hard to tell how it handles longer stories, keeping characters consistent, or intense action scenes.