Every camera, every phone, the whole shoot — exactly as it comes off the cards. One machine, overnight. Your footage never leaves it.
You hand it footage exactly as it comes off a shoot. Nested card dumps, mixed cameras and phones, thousands of files named C0053.MP4. It works out what happened, who matters, and what the film should be — and hands back a rendered film plus an editable Premiere or Resolve timeline.
Below is what it does and doesn't do, as a Q&A — easier to skim than a wall of text. Jump to whatever you care about.
Yes. No timecode needed — off the footage alone.
Yes. Also when two cameras spend three hours in different rooms and only overlap for a few minutes. Also when a single card drifts within itself.
Every recurring person, indexed across the whole event. Outfit changes, lighting changes, doesn't matter.
No. It works out roles on its own — anyone with enough footage to be identifiable gets one. The couple, the parents, the siblings, whoever the event makes prominent. There's no preset list it's matching against. And no names, no labels, no identity lookup against anything. Ever.
The footage, never — no frames, no video, nothing. Some text processing currently goes through an API for speed, but that's optional, not a requirement.
That's all I've tested it on — real multi-day Indian weddings — so that's all I'll claim. But nothing in the system is wedding-specific. No hardcoded list of ceremonies or roles anywhere. The structure, the stages, who matters — all worked out from the footage. In principle the same system runs on a Western wedding, a corporate event, a birthday. Just haven't run it on one yet.
Yes — grouped by the actual stages and moments of the event. Not by card and filename.
Optionally. Drop music files in the input folder and they get analysed for fit — it'll match them to the footage and use as much or as little as works. Drop in a brief.txt with guidance — including how long you want the film — and it takes that into account. Neither is required; it runs on footage alone.
Hours of it, on camera mics, in the way people actually talk at Indian weddings — Hindi and English switching mid-sentence.
No. Cuts land on phrase boundaries, and in/out points slide off any word they'd chop in half. You never hear a sentence cut off on playback.
It does, per clip. Decides where the source audio matters, ducks the music under it, brings it back after. That pass where you hand-keyframe the music down and up forty times a film? Gone.
Yes.
Yes. Not proportional to what was shot — proportional to what a finished film needs. The result has a film's proportions, not an archive's.
Yes. Hand it a 14-second take, it places the 2.4 seconds that play.
No. The bed gets edited like the picture does. A song enters from its right stretch, not its beginning. Sections continue instead of restarting. Cuts land on the music's beats, and fades are measured in the track's own tempo, not milliseconds — fading under a speech a quarter-beat off is the kind of thing you hear on every playback for the rest of your life. It doesn't do that. And it leaves when the film needs it to.
No. It revises early decisions in light of later ones instead of committing linearly.
Two things. A rendered film — watchable, complete, music mixed. And the same cut opened in Premiere or Resolve as a real editable timeline. Speed ramps and music bed intact, alternate takes stacked on disabled tracks so auditioning a different shot is one enable. The timeline links to your footage by relative paths — it sits alongside the input folder, so it opens with nothing offline, on whatever machine the drive's plugged into. The render shows what it decided. The timeline is where an editor takes over. Not a locked MP4 pretending to be one.
No. There's a shared Google Drive workflow — I own the Drive and pay for the storage. You upload the footage, and everything lands in a folder next to it: the film, the timeline, even the JSON. The timeline links to your videos by relative paths, so there's no duplicated footage — the input's already on your machine, so you just download the output folder (way smaller than the input), put it next to your footage, and it opens with nothing offline.
No, not today. The film comes out complete — synced, structured, scored, mixed — but the final taste and polish a client sees isn't there. An editor takes it from that timeline.
Not yet. And here's the thing — everything needed to answer that already exists. The entire event lives in one JSON file: every shot with who's in it, their roles, the stage of the event, what's said, expressions, a second-by-second account, quality scores. It falls out as a byproduct of deciding the edit. What doesn't exist is the search box on top. That's next.
I don't have rights to show client films publicly — so the demo is your own footage, run through the system. You get back the film, the editable timeline, and the full analysis JSON. You tell me where it breaks. That's the whole deal.
Email me about a free runNo footage, just thoughts or questions? Same address: hello@newcut.ai. I read everything.