The models

We don’t build models, and we don’t resell them.We direct them. Roll & Cut is a production studio that runs on other people’s models the way a film studio runs on other people’s cameras — what you are buying is the direction, the continuity and the edit, not access to any one of them.

What that means in a single film

One finished piece is not one model’s output. A writers’ room drafts the script and a director breaks it into shots. A camera model draws a reference sheet so the same face and the same room survive every cut, then composes each keyframe from it. A motion model animates those frames. Voices are cast and performed, a score is written to the finished length, and the whole thing is mixed and assembled. Those are different models from different companies, chosen per shot.

That coordination is the product. A single model, however good, cannot hold a character’s face across eleven shots, keep a narrator in sync with the picture, or cut a scene — and no amount of direct access to one gets you a film.

Who we use

Today we direct 29 models across 6 providers — Anthropic, ElevenLabs, Google, OpenAI, Runway and xAI — plus 79 voices. Broken down by the job each does, that is 9 for writing, 6 for images, 5 for motion and 9 for review.

These numbers change, and this page counts them rather than quoting them. We add a model when we have measured it against the one it would replace, and we drop one when it stops being the best way to do its job. We deliberately do not publish a fixed list of names: it would be out of date within weeks, and which model drew a particular frame is our problem to get right, not yours to research.

If you do want to choose

On most plans you never think about any of this — the studio picks, shot by shot, within the budget you set. On Pro and Studioyou can take the decision back and choose which model fills each role: the writers’ room, the camera, the motion model and the reviewer. That is a control over your own production, not an API key — you are still commissioning a film, not renting a model.

What we mark

Every file we deliver carries standard IPTC metadata marking it as AI-generated, with no way to switch it off. That is partly honesty and partly practicality: some of the models we direct watermark their own output regardless of what we do, and platforms increasingly expect the declaration.

The documentation covers how a production actually runs, and pricing covers what it costs.