Where it runs and where it works

Where does the engine run?

The engine runs on infrastructure we control inside the EU, and the files it works on are stored next to the machines that process them. The vision, speech and audio models are open-weight models, which means their trained weights are published, and they run on that same infrastructure. The engine sends no footage to an outside AI service.

Two parts of Everyframe still use providers outside the EU. One is the language model that plans the edit, and the other generates a missing shot. Both are used under the Standard Contractual Clauses, the contract terms the EU approves for sending personal data abroad. We plan to move both to providers that process data inside the EU.

Where it already works

The engine has run paid client work since spring 2026. For GZS's national innovation awards it made a 30 to 40 second video for each of 235 innovations in six voice versions. That is 1,410 videos, rendered unattended and shown at the live award events.

VARNOŠKA's 30 exercise routines have been live since August 2026, and the Faculty of Logistics hides faces and plates in its research footage with the engine. The computer vision chapter has both. Our own JT Cut, which turns a set of photos into a narrated video, runs on the same engine.

Everyframe itself has been open to invited users since September 2026. A 30-second ad renders in 11.9 seconds, which is about 400 ads an hour on one server.

What comes next

Every part of a video that is generated will carry C2PA content credentials and a watermark inside the file. C2PA is the open standard for recording, inside a media file, where it came from, and the watermark is what Article 50 asks for. The planning and generation models will move to providers that process data inside the EU.