See it survive
We’ll watermark a video, then attack it the way the internet does — and try to find the mark in every copy.
People can no longer tell what’s AI.
AI video is now photorealistic. Regulators (EU AI Act Art. 50, California SB 942) are starting to require that AI-generated content is labeled in a machine-readable way — and that the label is robust. Today’s labels fall off the moment a video leaves the platform that made it.
How it works
Mark
Your AI pipeline sends each generated video to Tectra (or runs our SDK). We embed a unique ID as a faint, invisible pattern in every frame.
Share
The video gets posted, downloaded, re-uploaded, cropped, mirrored and compressed — the mark comes along for the ride.
Check
Anyone — a platform, a newsroom, a viewer — checks a copy. Tectra recovers the ID and shows who made it and with which tool.
What it does — and doesn’t — do
- Survives: re-encoding, heavy compression, downscaling to 360p, frame-rate changes, cropping (keeping ≥ 60% of the frame), mirroring, brightness/contrast changes, and combinations of these.
- Not yet: rotation, extreme crops (e.g. 16:9 → 9:16 vertical), screen-recording at an angle. These are on the roadmap.
- Not a lock: like every watermark (including Google SynthID), a determined expert with access to the algorithm can remove it. It’s designed to survive everyday sharing, which is where labels get lost today.
- No false alarms: each mark carries a 32-bit checksum, so a random video is wrongly flagged roughly 1 in 4 billion times.
Building with AI video?
We’re onboarding a small number of AI video platforms and studios for early access to the API.