Glossary
Invisible watermark
An invisible watermark is information encoded into the pixels of an image rather than into its metadata.
Updated
A watermark of this kind modifies the image data in ways designed to be imperceptible but detectable, and to survive the ordinary things people do to images — cropping, resizing, re-compression, screenshotting.
The distinction from metadata is the one that matters. Metadata sits beside the picture and can be removed without touching a pixel. A watermark is in the picture, so no metadata operation affects it.
This is why stripping a content credential does not make a generated image undetectable. The label goes; anything embedded in the pixels stays, as do the statistical traces classifiers look for.
SynthID vs C2PA compares Google's pixel watermark with signed credentials directly.
Related terms
- Provenance — Provenance is the record of where an image came from and what has been done to it since.
- C2PA — C2PA is a provenance standard that attaches a signed, tamper-evident manifest to a file, describing its origin and edit history.
- Re-encoding — Re-encoding is decoding an image and compressing it again, which discards detail on every pass through a lossy format.