Signature module · JT Digital
Everyframe
Video work that never leaves infrastructure we control.
Detect · Read · Redact
Everyframe is our own engine for video and computer vision. It composes video, finds faces and licence plates, reads what is on screen, blurs what needs blurring, and restores footage that would otherwise need a specialist studio. Every model runs on machines we control in the EU. Your footage is stored in the same datacentre as the machines that read it, so material that cannot leave your control can still be processed.
What it does
Six jobs that usually need three vendors.
All of it one engine, on the same machines, with one bill.
Hundreds of videos, no editor
A spreadsheet of data becomes finished, on-brand video: stills that move, titles, captions, watermarks, music that ducks under the voice.
Find every face and plate
Faces, licence plates and eighty classes of object, found across an archive nobody has time to watch, with an identity held across frames.
Blur that cannot be undone
Blur, pixelate or black-box, burned into the export, so footage can leave the building without leaking who was in it.
Text and speech out of footage
On-screen text in eleven languages, and spoken words with word-level timing, so an archive becomes something you can search.
Old footage made usable
Denoise, deblur, upscale and fill in missing frames, for material that would otherwise be written off.
A sealed receipt per export
Every export carries a record of what the pipeline did to it, checkable offline, with no account and without taking our word for it.
Why it matters
Four things you get back.
No middleman in the pipeline
The engine, the models and the applications are built in one place. There is no third-party AI service in the path. The model weights sit in storage we control and inference runs on the same machines that hold the footage. Nothing about your material becomes a vendor's business, and nothing in the chain waits on someone else's roadmap.
Weights are pinned by checksum, so tampered weights fail the job before they produce a result.
Even those weights come from that same storage, so nothing is pulled off a third party's server while your job runs.
No per-frame vendor fee. The cost follows our capacity, and no outside price list can move it.
Your footage never goes anywhere else
Video is the heaviest and most exposing thing a client hands over, so it goes exactly one place. Files land in a bucket we control, in the same datacentre as the machines that decode, detect and render them, and the export comes back to that same storage. There is no file-sharing service in the middle, no vendor upload endpoint, and no copy of your material resting in somebody else's account.
The queue carries a reference to the job. What the frames contain never rides through the message broker.
Storage, broker, GPU and encoding machines all sit in one datacentre, so a file crosses between them in a fraction of a millisecond and never over the public internet.
Footage stays in the EU, and trains nothing
Storage, detection, rendering and export run in one European datacentre. Footage is never used to train or fine-tune models. That commitment is written into the contract. Where a project cannot touch a cloud at all, the same pipeline runs on your own server, or fully offline on a small board with no network interface.
Retention is per project. Raw material is purged on the schedule you agree, and the schedule is in the agreement.
Residency and non-training are the claims we make. We do not decorate them with certification badges we have not earned.
You can check the work yourself
Each export carries a sealed receipt describing what the pipeline did from ingest forward. It verifies with a standalone tool, offline, with no account and no connection to us. A data protection officer gets evidence they can check themselves. The receipt covers what happened after ingest. It does not authenticate the camera original, and we will not pretend otherwise.
In practice
Where it already runs.
About 250 branded videos a year
The national innovation awards platform renders its videos on this engine. What used to be a fortnight of manual voice-over and editing per video is now a render measured in seconds, across ten Slovenian regions.
Footage that had to leave the building
Recorded footage the theatre had to hand over, with every uninvolved person unrecognisable. The engine found the faces across the recording and the blur was burned into the export, so the copy that left could not be turned back.
Thirty exercise routines, counted by the engine
A workplace safety library for the Koroška region. The engine finds the person in each clip, reads the pose, counts the repetitions, then composes the routine: title cards, a burned-in counter, Slovenian narration and music ducked under it. Thirty routines out of 126 source clips, no editor involved.
First contract signed for anonymised video
An agreement covering 480 hours of urban footage: every identifiable face and readable plate anonymised, processing kept inside the EU, no training on the material, and raw footage purged after acceptance. Those terms are in the contract.
JT Cut and Huluma run on it too
jtcut.com turns photos into a narrated video. huluma.eu finds every face in a family photo and covers it before the picture is shared. Both sit on this engine, which means we meet its failures before a client does.
The rest of the head start
Two more we own outright.
Next step
Have footage that cannot go to a cloud?
Fifteen minutes is enough to say whether this fits, what it would cost, and exactly where the work would run.