The idea in one paragraph
A camera or an application writes a small signed record into the file. The record names the producer, the software, and each edit step. A viewer with the right software checks the signature and shows the chain. The C2PA specification defines the format, and Adobe brands its implementation Content Credentials.
What a manifest can hold
The exact contents depend on the producer. A manifest commonly carries some of the following.
- Origin The device, the application, or the generative model that produced the asset, plus a claim about whether the source was a camera capture or synthetic media.
- Edit history An ordered list of actions such as crop, colour adjustment, or composite, each attributed to a step in the chain.
- Identity The signer's certificate, and optionally a creator name or an account. This is the field that surprises people.
- Assertions about AI use A declaration that a generative model made or altered the asset, and sometimes a declaration that a watermark was applied.
Vaultaire reports a declared watermark action when a manifest states one. That is a statement inside the record, not a pixel inspection, and the report labels it that way.
Why the record usually goes missing
A manifest travels in the file. Anything that rewrites the file can drop it, and most things rewrite the file.
| Action | Manifest survives? |
|---|---|
| Copy the file to another disk | Yes |
| Take a screenshot of the image | No, the screenshot is a new file |
| Upload to a social platform | Usually no, uploads get re-encoded |
| Send through a messaging app | Usually no |
| Export from an editor | Only when that editor supports C2PA |
| Run a metadata remover | No, removal is the point |
So a missing manifest carries no meaning. Its presence tells you something. Its absence tells you nothing, which is the same asymmetry that governs checking whether an image is AI generated.
When removing the record is the right call
Provenance helps a news photograph and hurts a private one. The decision comes down to who reads the file next.
Keep the manifest when a viewer needs to verify the origin: press work, evidence, licensed stock, anything where the chain is the value. Remove it when the file is going to a client, a forum, a marketplace listing, or a public post and the chain would publish your software, your account, your edit history, or a prompt you wrote.
The removal itself is plain. Load the image in the metadata tool, read what the report found, then remove. The tool deletes identified C2PA containers together with XMP and IPTC provenance fields, writes a separate clean copy, and scans that copy to confirm the fields are gone before the download unlocks.
What this does not do
Removing a manifest removes a record. It does not remove a pixel watermark, and it does not verify a signature for you. Vaultaire reports the container and its declared contents. Checking a certificate chain against a trust list is a separate job that needs the signing infrastructure, and our tool does not attempt it.
For the difference between metadata removal and pixel watermarking, read how to remove an AI watermark from an image.
Content Credentials FAQ
What are Content Credentials?
A signed provenance record attached to a media file. The record, called a manifest, says which device or application produced the file and which edits followed. C2PA, the Coalition for Content Provenance and Authenticity, publishes the specification that defines the format.
Is C2PA the same as a watermark?
No. A C2PA manifest is data attached to the file and it comes off. A pixel watermark changes the picture itself and survives metadata removal. The two solve different problems and often appear together.
Should I remove Content Credentials from my images?
It depends on what the manifest holds. It can carry your name, your software, a prompt, an account identifier, and a full edit history. Remove it when that detail should not travel with the file. Keep it when you want a viewer to verify the origin.
Does removing a manifest damage the image?
No. Vaultaire deletes the container without re-encoding the picture, so the encoded image payload stays byte-identical and the visible photo does not change.
Why do most images have no Content Credentials?
Adoption is partial and the record is fragile. Screenshots, social uploads, messaging apps, and many editors write a new file and drop the manifest. A missing manifest is normal and proves nothing.
Can the tool verify the signature?
No. It reports the container and what the record declares. Validating a certificate chain against a trust list needs the signing infrastructure, which sits outside a local browser tool.