How a file gets Content Credentials
C2PA is a coalition whose open standard, branded Content Credentials, is meant to help people trace where digital media came from and how it was changed. A capable camera, editor or generator writes a manifest into the file and signs it. Any later program can read the manifest, check the signature and show the history, which is the whole design in a sentence.
What is inside a C2PA manifest
In the C2PA vocabulary, a manifest bundles assertions, a claim and a claim signature, and each piece does a different job.
- Assertions are statements: that the image was captured by a device, that an editor applied certain actions, that a generative model produced it, who the creator is.
- The claim lists those assertions and the information needed to bind them to the file. It's the thing that gets signed.
- The signature is made with the signer's private key, so a validator can tell who signed and whether the claim was changed.
- A hard binding is a cryptographic hash of the asset. It lets a validator confirm the manifest belongs to this file and that the file hasn't changed since.
In a JPEG, the manifest travels in APP11 segments using the JUMBF box format. Other formats keep it in their own containers, and in a PNG it's in a chunk named caBX. The metadata viewer on this site looks for those containers.
What a valid credential proves and what it leaves open
The C2PA specification describes what it records as provenance and integrity, and makes no claim about truth. A valid credential says the signer made these claims and that the file matches them; whether to trust the signer remains your call. A camera maker signing a photo is a strong signal, an unknown signer is a weaker one, and a perfectly valid credential on a staged photo is still a staged photo.
Where Content Credentials go when you share an image
Most images you meet have no credentials, and some had them and lost them. The manifest lives in the file wrapper, so any process that rebuilds the file from pixels drops it: screenshots, many editors' save paths, format conversions, and services that recompress uploads. The specification mentions soft bindings, such as fingerprints or watermarks, as a way to match a stripped asset to a manifest stored elsewhere, but version 2.0 of the specification defines no soft-binding algorithms. In practice, absence is the normal state, which is why a missing credential tells you so little about whether an image is AI generated.
Checking a file for a C2PA marker
The viewer on the photo metadata page tells you whether a C2PA marker is present and prints a few readable strings from it: the claim generator, assertion labels such as c2pa.actions, and any IPTC digital source type. It doesn't validate the signature or the hash. For validation, use a verifier built for it, such as the Content Authenticity Initiative's Verify site, and read its privacy terms before you hand it a file.
Reasons to publish a copy without the manifest
There are ordinary reasons to share a copy without the record. The credential may name the device or software you used, a location, or an account you didn't mean to expose. A journalist protecting a source, or someone posting family photos, may prefer a clean file. The cost is that the copy carries no proof of origin, so anyone who wants to check will have to take your word. Removing your own credentials from your own images is a privacy choice. Removing credentials doesn't turn an AI image into a photograph, and this site does not help anyone present one as the other.
What stripping the manifest leaves behind
Removing a manifest by re-encoding doesn't target the pixels, so a visible mark or an invisible watermark in the image stays. The AI watermark guide explains the difference between the two.
Questions people ask
What are Content Credentials?
A signed provenance record attached to a file, built on the C2PA standard, stating how the file was made and edited. It's data stored beside the picture, not a change to the picture.
Is C2PA a watermark?
No. A manifest is separate data stored in the file wrapper, while a watermark is a signal embedded in the pixels. They fail in different ways: one disappears on re-save, the other can survive it.
Should I strip Content Credentials from images I publish?
It depends on what you want to protect. If the record exposes details you'd rather keep private, strip it from the copy you publish. If you want to prove where the image came from, keep it.
Does stripping a manifest damage the image?
Removing the wrapper doesn't change the picture. Re-encoding to do it does recompress JPEG and WebP slightly, so choose a high quality setting and compare the copy with the original.
Why don't most images carry Content Credentials?
Few cameras and tools write them, and the ones that do often lose them in editing, sharing and recompression. For most of what you see online, no record was ever attached or it didn't survive.
Can the viewer verify the signature?
No. It reports that a marker exists and prints readable strings from it, such as the claim generator. Use a dedicated verifier for validation, and read its privacy terms first.
Sources