VideoMaMa is a model that turns simple black-and-white object masks into soft, precise cutouts (alpha mattes) for every frame of a video.
Saber is a new way to make videos that match a text description while keeping the look of people or objects from reference photos, without needing special triplet datasets.