VMCP gives an AI agent a complete editing suite — cut, framing, captions, sound, effects — and the senses to check its own work. It edits the rushes you filmed. Nothing is generated.
An agent cannot watch or listen. VMCP turns the edit into text it can reason about — what is on screen, where the faces are, what is said, where a cut lands in the waveform — and shows real frames only where it matters.
Open tools, not templates: the agent writes its own components, picks its own fonts and decides every parameter from what it measured.
Fillers, retakes, pauses marked on the transcript. Every cut checked against the waveform and pitch; flashes of baked-in shots caught frame-accurately; ripple fixes.
Face-driven punch-ins with a real focus, steady locks, animated reframes, focus blur around the speaker, person matte for text behind the subject.
Captions and motion graphics as code, in the channel's own design system, placed where the frame has room — never over a face.
Background noise analysed in the pauses, removed with a targeted filter that learns the room's noise print, verified band by band.
Recorded sound effects the audience knows, described in text — peak, attack, envelope — so they land exactly on the cut.
Any image or clip as a layer: chroma key, person cut-out, outline, blend modes, animation. Light leaks, film burns, emoji, reactions.
Tempo, bars and energy sections read from the track, so the downbeat lands on the key cut and the drop on the reveal.
vmcp map reads the edit event by event; vmcp frames renders real frames exactly where a doubt remains.
Sound effects, light leaks, film burns, dust, stickers and emoji that editors actually use — recorded by people, licensed CC0 or CC BY, hosted here so every agent gets the same files.
Open the library