Vibemotion
A video editor where humans and AI work on the same timeline.
- Views
- 116,185
- Bookmarks
- 1,226
- Likes
- 1,019
- Waitlist
- 5,000+
- Daily active users
- 100+
Vibemotion is a multi-track video editor for clips, text, images, shapes, sound and motion graphics written as code. Every change to a project is one of a small set of typed actions, such as adding, updating, splitting or deleting an item on the timeline, and the interface and agents connected over the Model Context Protocol (MCP) submit the same actions to the same live project. An agent can therefore make any edit a person can. It can also render any frame or range of the timeline and look at the result before its next action. Vibemotion is in private beta at vibemotion.ai.
seq 36
source client
timestamp 1771604021
mutation {"action": "update-overlay", "overlayId": 12, "updates": {"from": 96, "durationInFrames": 45}}
result {"id": 12, "type": "text", "row": 2, "from": 96, "durationInFrames": 45, …}
mcp), open squares are actions from a person in the editor (source client). The curve joins the agent adding text overlay 12 to the person retiming it 14 entries later; that entry is printed above, with its result abridged.Each of these actions is appended to the project's log with its arguments, its result, its source and a timestamp. A finished project is a trajectory of how its video was made, edit by edit, and because the source separates the interface from an agent, the log also marks the points at which a person changed what an agent had made. We are building RL environments for long-horizon video editing from these logs: the state is the project and its media, the action space is the editor's own, and the demonstrations and corrections come from people editing their own work with their own taste.