← All stories
discussionr/museagent · Nov 16, 2026

AI motion videos with Muse

A user testing animated video styles found Claude nails them from one prompt plus a reference, but Muse produced ugly results from the same instructions, even with a repo to follow. The likely gap is prompting, not the model: split the video into layers — static background, one moving element, one pulse effect — describe the reference frame by frame, composite with moviepy or ffmpeg, and keep each clip to a short loop.

Read the full source →