Editing is judgement. Judgement can be learned.

01The problem.

Generative video models answer one question: what should these pixels look like? An editor answers a different one: given these pixels — this rambling take, this restart, these forty minutes of a morning that actually happened — what matters, and in what order? That is a decision problem over real footage, and it is nearly untouched. The tools that exist automate mechanics: silence removal, transcript search, templates. Useful. But mechanics were never the hard part. A cut is a claim about what matters. Nobody has trained a model to make that claim well.

02The data no one else has.

RCM 1’s core training data is narrated editing: a professional editor working on real projects in a capture tool we built, explaining every decision out loud as he makes it. Why this take and not the other six. Why the cut lands on this frame. Why the pause stays in. Each session yields the footage, the finished cut, and the reasoning that connects them — the state, the action, and the why. Most editing data in the world is before/after. Ours is before/after/because.

—Pixels vs. decisions.

The internet is filling up with AI that makes a fake you — a cloned voice, a synthetic face, words you never said. Real Cut does the opposite. Its editing pipeline never generates a frame: it cuts the footage you actually shot, so the person on screen is the real one, on their best take. That is the whole point. Real you, elevated.

The hard part of editing was never cutting. It’s knowing whichcut — which take carries, which line to keep, where the pause earns its place. We’re training RCM 1 on the one thing no one else has: professional editors narrating that decision out loud, on real footage, take by take. Before, after, and because. So when it hands you a short, there’s a reason behind every cut — not a template it filled in.

03Provenance.

Everything RCM 1 trains on is ours to train on. The founder’s own edits and footage. Licensed professional material, paid for. Sessions from the editors who teach it — paid, credited, named in the model card. Public-web material is used only as reference for measurement — calibrating what real professional editing statistically looks like — never as training pixels. Nothing scraped. We hold this line because the alternative poisons the well: a model of taste built on theft has already failed the taste test.

—The editors who teach it.

The taste in RCM 1 belongs to working editors. They edit real projects with a microphone on, narrating every decision, and they are paid for every session, credited by name, and named in the model card. We don’t audition people for free. If you edit for a living and want to teach it, write — pick one cut in your reel and tell us why you made it.

ZAN@REALCUT.AI

04Evaluation.

There is only one honest benchmark for an editing model: the cut a professional would have made from the same footage. We maintain a holdout set of finished professional edits the model has never seen, and we score RCM 1’s cuts against them — selects, structure, and frame-level cut placement. We have not published numbers yet, because the numbers are not yet worth your time. When they are, they’ll be on the model page, with the methodology, including what the model still gets wrong.

05What RCM 1 is, and isn’t.

RCM 1 proposes; you decide. It hands you a finished short with a reason attached to every cut — and when you ask for a change, in plain words, your corrections are exactly the signal it is designed to learn from. It is not a video generator. It will not make footage of things that didn’t happen. It is not a replacement for editors; it is trained by them, credited to them, and pointed at most of editing — the part that is finding the cut — so that the people with something to say can spend their energy on saying it. We think the difference between those two futures is the most important product decision in this field. We’ve made ours.