Captions that hit on the beat.
Auto-transcribed and word-synced, with karaoke highlight, clean wrapping, and your brand styling — every line landing exactly when it's spoken.
The AI-native editor for talking-head video. Describe the outcome — the agent cuts, places b-roll, writes captions and scores it, on a real multitrack timeline you can still edit.
Not a mockup — a screen capture of a single agent run, start to finish.
One frame, both states. Drag the divider to the left and the mechanical pass appears — trimmed, captioned, graded, scored.
Auto-transcribed and word-synced, with karaoke highlight, clean wrapping, and your brand styling — every line landing exactly when it's spoken.
The agent reads the transcript, watches your clips, and lays b-roll where the words earn it — then puts a transition at each in-point and a whoosh on top of it.
WebCodecs renders locally — no upload, no render farm, no desktop install. Reframe once and export every aspect ratio your channels need.
Nothing you upload trains any model. The agent reads your project only to make the edits you ask for.
Exports run locally over WebCodecs — footage is never shipped off to a render farm.
Media lives in your own private storage bucket — never shared, never sold, deleted when you say so.
One undo reverts an entire agent turn. Nothing is flattened, nothing is locked.
Plans open at launch. Move up only when Lumiere is already saving you more time than it costs.
For getting real edits done without the daily beta ceiling.
For regular AI edits, captions, and assisted finishing with fewer beta limits.
For heavier motion, deeper analysis, and larger multi-step edit sessions.
Lumiere takes the mechanical pass. You keep the final cut.