Covert watermarking

A covert watermark writes a short identifier into the fabric of an image — invisible to the eye, but recoverable with your secret key even after the image has been passed around. It's how you later prove "this picture came from me".

Embed

In the Embed panel:

1. Drop in the image you want to mark.
2. Key (required) — your secret. The same key is needed to extract or detect the mark later, so treat it like a password and keep it consistent across your images.
3. Text ID — up to 32 bytes: a name, order number, client ID, URL slug.
4. Strength (optional) — how firmly the mark is written. Higher survives more abuse (recompression, mild edits) at a slightly higher risk of visible artefacts in flat areas; the default is balanced.

Download the marked image and publish that instead of the original.

Extract

Drop a marked image into Extract with the same key and the embedded text ID comes back. Use it when you need to know which mark an image carries — say, identifying which client leaked a preview.

Detect

Detect answers a simpler question — is my mark present? — and returns a confidence score (e.g. Detected, confidence 0.994). It's the one to reach for when scanning images in bulk, and it's also available through the developer API.

How well does it survive?

The mark is designed to survive what images normally go through online — format conversion, recompression, moderate resizing. It will not survive everything: heavy cropping, aggressive filters or a screenshot of a tiny thumbnail can degrade it below detection. When robustness matters, embed at higher strength and test your own pipeline: mark an image, put it through the treatment you expect, then Detect.

Covert watermark vs steghide

Both hide data invisibly, but they're built for different jobs: steghide maximises payload (whole files) and assumes the image travels intact; the watermark maximises survivability of a tiny ID. Don't use steghide for provenance or the watermark to smuggle files.