The problem
AI makes it cheap to fabricate video of a newsroom, a candidate, an executive or an ordinary person. Logos, faces and voices can all be copied. Corrections arrive after the damage is done, and detection tools are a step behind the newest model.
Don't chase every fake. Make the real thing provable.
How it works
01Bind at publication
Saivd embeds an invisible, encrypted watermark in every image and video frame, bound to the source's account.
02Verify in the player
The Saivd player checks the file. Its live QR refreshes every second, so a screenshot of it won't verify.
03Altered won't verify
Alter the file and it no longer verifies. Only the original passes, and media you never signed won't verify as yours.
04Fits your workflow
Saivd runs inside your workflow, after your transcode, and integrates with your existing publishing and playback systems.
The check runs on the file as you released it. That strictness is the point: viewers see exactly the file that was signed.
Precise by design
Identity, rights and truth are separate claims. Saivd makes one of them provable and doesn't pretend to make the others.
What Saivd confirms
What it doesn't claim
Patented. Patented video technology, US 12,348,649.
Video and still images. One approach for both.
Fits your systems. Integrates with your existing publishing and playback systems.
Complementary. Works alongside C2PA, fact-checking and deepfake detection.
Use cases
Standards
Saivd embeds the source's identity in the media itself as an invisible watermark. It doesn't yet read or write C2PA labels or hold C2PA certification. Registering as a C2PA soft-binding method is being evaluated.
US Patent 12,348,649, granted July 1, 2025, covers embedding an authentication signature in video.
Supports transparency obligations like EU AI Act Article 50, alongside your detection and moderation tools.
FAQ
Yes. Saivd embeds an invisible, encrypted watermark in every image and video frame, bound to the source's account.
Yes. Saivd watermarks still images as well as video. Our patent covers the video method.
Yes. Every column of a frame feeds the check, so changing it breaks verification. Every frame is covered.
No, and that's deliberate. Re-encoding changes pixels, which is what the check catches. So we integrate after a platform's own transcode, and the file viewers see is the file that was signed.
No. A deepfake won't carry your signature, so it won't verify as yours. Detection tools estimate; we check a signature. The two work together.
No. It proves which account published the media.
Your existing identity process. Saivd binds media to the account you've already verified.
A screenshot won't verify, because the QR refreshes every second. And the real proof is the player's check of the file itself, not the picture of the QR.
No. We're complementary to C2PA: the source's identity is embedded in the media itself, and we're evaluating registration as a C2PA soft-binding method.
No. Saivd watermarks the video frames. Audio isn't covered today, so a swapped soundtrack may not be caught.
Not for video today. Keep your masters; Saivd watermarks the copies you release.
Yes. US patent 12,348,649, granted July 2025, covers embedding an authentication signature in video.
Each pilot is built around your needs. We'll scope it with you on a 20-minute call.
Saivd verifies who published a video or image, never whether its claims are true. Complementary to C2PA and deepfake detection.
We're partnering with a select group of industry leaders to help shape how source authentication fits into their everyday workflows.