Describe the cut. Take the cut.
vidoflip is an agentic video editor. Tell it the edit you want, in plain language, and an agent performs it on a real timeline: numbered operations you can read, candidates you can compare, and a cut you can still open and change yourself.
vidoflip_get_project_snapshotdone
30.7sMusicians on beach at sunset
15.3sMotorbike through rice paddies
69.9sScenic road, rocky cliff cut
9.2sSunset over the ocean
14.4sUnclear — need to review
YouReview the unclear one.
vidoflip_grab_frames · vidoflip_get_transcriptdone
It’s a spoken intro. Use it to open, start the music when it ends.
vidoflip_submit_plan kind=sequence kind=musicdone
The edit is a list of operations, not a black box
If you have watched a coding agent work through a plan, you already know the shape: an instruction goes in, legible operations come out, and nothing happens that you cannot read, revise, or reject. vidoflip brings that working loop to footage.
It watches the footage
The agent analyzes what you gave it (scenes, transcript, faces, beats) before it touches a single frame. Any footage: screen recordings, phone video, a talk, an event.
It operates a real editor
Cuts, reframes, captions, titles, music: all performed as editor operations on a multi-track timeline, in order, each one visible in the session log.
You take the cut
The agent proposes candidates. You compare them, take one, tweak anything by hand in the same editor, and export. The pick is always yours.
One agent, a working editor’s toolbox
Everything below ships in the product today. Each one is an operation the agent can perform and you can inspect.
Pace recut
Tighten a rambling take to a target length. Shorten-only, structure preserved.
Beat-synced cuts
Cuts land on the music’s detected beat grid, or an existing cut re-times to it in place.
Captions that follow
Auto-captions anchored to the content. They slide with every later edit instead of drifting.
Agent-made titles
Motion titles and lower-thirds authored per request: typeset, animated, legible on any footage.
Music that ducks
A music bed with a start offset and automatic ducking under every spoken line.
Smart reframe
Reframe wide footage for a different aspect with on-device face detection guiding the crop.
Footage analysis
Scene detection, frame grabs, and transcription give the agent (and you) a map of the material.
Candidates, versioned
Every proposal is a saved version. Compare, take one, or ask for another round.
It proposes. You dispose.
For one instruction the agent returns competing candidates, each a readable list of operations over your footage. Taking one is a single act. So is rejecting all three.
- Length0:58
- Cuts14, beat-snapped
- Opens onthe demo
- Length0:45
- Cuts11, beat-snapped
- Opens onthe punchline
- Length1:12
- Cuts9, wide pacing
- Opens onthe intro
Renders on your machine
The desktop app drives native FFmpeg on your hardware; the web editor renders in your browser. Your footage is not parked on our render farm. We don’t have one.
A general editor, on purpose
Not a clips tool, not a template mill. Any footage in, any shape out. The agent works from what your material actually contains.
The desktop app is in closed testing. Get in line.
- First access when we open the doors
- Free Pro plan for early members
- A direct line to shape the product
We’ll write when your invite is ready. No newsletter, no drip. Just the invite.
We use your email for the invite and nothing else.
What footage does it handle?
Any. Screen recordings, phone video, talking heads, events. The agent reads the material first and works from what is there.
What platforms?
macOS desktop first: the full agent-plus-editor product. A web companion follows.
When do invites go out?
Rolling, as closed testing expands. Early-access members are first, in order.