5 min read

AI Watermarks vs Metadata

Embedded pixel signals and EXIF/C2PA manifests solve overlapping problems in different ways.

Metadata: easy to read, easy to strip

EXIF, XMP, and C2PA fields sit in the file header. Any image viewer or hex editor can read them, but a single re-save in most apps wipes them out.

Pixel watermarks: harder to add, harder to remove

Pixel-domain watermarks survive a screenshot, a re-encode, and moderate editing. But adding them requires a trained model and detecting them requires a paired key — they're not human-readable.

They're complementary

Robust provenance combines both: cryptographic metadata for verifiable claims when the file is intact, plus a pixel watermark as a fallback for screenshots and re-uploads.

Try the tool

Check AI Watermark

Upload an image to scan for AI watermarks, provenance signatures, and metadata clues.

Open Check AI Watermark

Frequently asked questions

Primarily metadata and known provenance strings in the file header. Pixel-domain detection requires vendor-specific detectors we don't host.
Back to Check AI Watermark