Image-to-video with a GLB instead of an image. Block the shot in headless Blender - character, pose, camera - render per-frame control sequences, and the video model paints life over it: film, cutscenes, character performance, any footage, with one persistent main character.
