Attack it yourself
Take the watermarked photograph below, do whatever you like to it, and upload the result. Blurry reads the invisible ID back out and names the copy it came from.
1 — Take the sample
This is a real Standard-watermarked file, issued to one recipient. Recompress it, brighten it, screenshot it, put it through a chat app and save it back.
issued to: a7f31c9e-4b02-4d85-9f6c-2e1d7b8a3c40
2 — Upload what you did to it
The image is decoded in memory and discarded — it is never written to disk, never stored, and never associated with you. That is the same guarantee the product makes, so the demo of it should visibly keep nothing.
Drop your attacked copy here, or pick a file.
What this detector reads
This page reads the Standard watermark, which survives recompression, colour changes and re-encoding, but not a crop, a resize or a screenshot — its detector keys on the image's exact pixel dimensions. Being specific about that is the point: you should be able to find the edge yourself rather than take our word for where it is.
The AI method does survive all of those. You can see it done to the same photograph, attack by attack, with the real recovered IDs, on the robustness demo.