Per-fan watermarking explained

Forensic vs ordinary watermarks

An ordinary watermark says “this is mine”. A forensic, per-fan watermark says “this copy was delivered to this account”. The first helps with ownership; the second helps with attribution. How to watermark OnlyFans content compares the kinds side by side.

How a mark goes in

1. Each fan gets an ID. A number, not a name. In Trace you type the fan’s username; Trace assigns the number.
2. The ID is encoded with error checking. Extra bits let a damaged reading be checked, so a garbled read is rejected instead of turning into a different fan’s number.
3. It is spread through the content. For video, across many frames and several windows of time. For photos, across the image. For voice notes, through the audio. Small changes, everywhere, at a level people don’t notice.

Because the mark is everywhere rather than in a corner, cropping a corner or covering part of the picture doesn’t remove it.

How a leak is matched back

1. You upload the leaked file or paste a link.
2. The mark is read. If the leak was cropped, rotated or screenshotted, it is lined up against your original first, then read again.
3. The reading is checked against your fans. It must read cleanly, belong to one of your fans and, for video, beat the next-best match by a set margin.
4. The result is either a match (“this copy was delivered to account X”) or inconclusive.

Why “inconclusive” is a feature

A system that always gives an answer will sometimes give the wrong one. Naming the wrong fan costs you a real customer and your credibility with the platform. So when the evidence is weak, a careful system names nobody. Inconclusive doesn’t clear any fan; it means the copy doesn’t carry enough to say.

Mass messages work slightly differently

For a mass message, you can’t prepare 800 copies by hand. Instead the content is built from short segments of two marked versions, arranged in a different pattern per list or per recipient. A leak is matched by its pattern, and a match must be less likely than one in a million to happen by chance. Details in how to trace a mass message leak.

What it survives, and what it doesn’t

Built to survive: re-encoding, downscaling, social re-uploads, crops, filters, mirroring, screenshots, screen recording software and short clips. Known limits: AI regeneration can remove the mark, several fans combining copies hides which one leaked, very heavy crops leave too little, and a phone filming a screen often won’t name anyone. We publish every test, misses included, in the robustness report.

What it can’t do

• Stop someone leaking. It changes what happens after.
• Trace content sent before you started marking.
• Remove leaked copies. That is a takedown job: DMCA vs attribution.
• Tell you who pressed upload. It tells you whose copy it was.

Words you’ll see

• Roster: the list of fans you have issued marked copies to.
• Gate: the checks a reading must pass before anyone is named.
• Registration: lining a leaked copy up against your original before reading it.
• Collusion: several fans combining their copies to hide the source.
• Control: an unmarked copy put through the same test, which must name nobody.

For a longer look at invisible marks versus visible logos, see forensic watermarks for creators.