A Content Credential is a signed note in the file.It is not a mark in the pixels.
C2PA Content Credentials are a signed note attached to a file, not a pattern in the pixels. Labs write them so a later reader can see that a model was involved. Gifi can strip the embedded credential from images and documents. Soft binding can survive that strip and re-link to a remote manifest.
Reviewed by Can Balkaya
Published
Process
How to verify the job
01
Inspect
Drop the original file into Gifi to list any embedded C2PA container before uploading it.
02
Strip
Continue to remove the local manifest and the metadata container that carries it.
03
Verify
Inspect the downloaded file again and keep the action list as the record of what changed.
Scope
Supported and unsupported surfaces
| Surface | Gifi action | Verification limit |
|---|---|---|
| Named metadata or invisible Unicode | Inspect, remove, and report named actions | Reinspect the returned bytes independently |
| Statistical patterns in ordinary words | Rewrite the prose | No certificate against a private detector |
| Visible logos, pixels, audio, or video | Unsupported | No removal claim |
A signed note, not hidden ink
C2PA — the Coalition for Content Provenance and Authenticity — defines a way to attach a cryptographically signed history to a file. The usual public name is Content Credentials. The record lives in metadata: an APP11 / JUMBF segment on a JPEG, a C2PA chunk on a PNG or WebP, or a manifest stored with a document.
Nothing in that record has to change the picture you see. The pixels can be identical before and after the credential is written. That is the opposite of a statistical text watermark, which lives in the words themselves.
What stripping actually does
The file cleaner deletes the embedded manifest and the metadata containers that carry it. After a clean strip, opening the file in an ordinary editor will not show a Content Credential. That is a verifiable fact: the bytes that held the note are gone.
It does not change how the image was made. It does not prove a human took the photograph. It does not speak to a statistical watermark in nearby text.
Soft binding can outlive the strip
Some C2PA deployments keep a fingerprint of the asset and a copy of the manifest on a remote service. After the local label is removed, that fingerprint can still match and re-attach the history. The industry name for this is soft binding.
Gifi can promise the local credential is gone. It cannot promise a remote registry has forgotten the file. Anyone who says a metadata strip “erases C2PA completely” is skipping this part.
Questions this usually raises
- Is C2PA the same as a text watermark?
- No. A statistical text mark such as SynthID-Text is a keyed pattern in word choice. C2PA is a signed metadata record on a file. They need different tools.
- Will colours or quality change?
- Not for PNG, JPEG or WebP. Those are stripped losslessly. ICC colour profiles are kept so colours do not shift.
- Can you remove a watermark baked into the pixels?
- No. Pixel-embedded image watermarks are out of scope. Audio and video are not supported.