What is C2PA metadata?
C2PA, shown to users as Content Credentials, is an open standard for attaching a signed provenance record to a file. The record states who made the image and with what tool, and it is bound to the content by a hash, so edits are detectable. It can be removed, and it never proves that an image is true.
Last updated: 2026-10-08
The short version
C2PA stands for the Coalition for Content Provenance and Authenticity, the group that publishes the specification. Its goal is to let a file carry a record of its origin that software can check. When that record is displayed to people, the feature is called Content Credentials. The record itself is called a C2PA Manifest.
What a manifest contains
The explainer describes a manifest as containing one or more assertions. Assertions are statements, such as which tool produced the image, which actions were applied to it, or that it was made by a generative model. They are gathered into a claim, and the claim is digitally signed by whoever creates the manifest.
A manifest is stored in a JUMBF structure, a general-purpose box format. ExifTool reads that structure and lists it as C2PA for JPEG, PNG, TIFF-based files and other formats. In a JPEG it travels in application segments, which is one reason ordinary image libraries do not show it: they read the segments they know and ignore the rest.
Why it is called tamper-evident
The manifest includes a hash that represents the content at the time it was created. The explainer calls this the hard binding. If someone changes the pixels after signing, the hash no longer matches, and a validator can say so. In the explainer's words, this means the provenance data can be said to be tamper-evident.
That wording is exact and worth reading slowly. The manifest does not stop edits and does not stop removal. It makes an edit visible to anyone who checks, if the manifest is still attached.
Why it can still be removed
Because the record travels inside the file, anyone can take it out. The explainer says so directly: asked whether provenance metadata can be removed, it answers yes. Ordinary processing does it by accident. The IPTC organisation points out that uploading to social media is one of the main ways metadata gets stripped, intentionally or not.
The specification's answer is the soft binding. A soft binding links a piece of content to its manifest through an invisible watermark or a fingerprint lookup, so a manifest kept in cloud storage can be found even when the file lost its copy. The explainer pairs the hard binding with a soft binding for what it calls durable credentials. Whether a given image has one depends on the tool that made it.
What it proves and what it does not
A valid manifest tells you who signed a claim and that the content has not changed since. That is useful for newsrooms and for tools that record edits.
It does not tell you that the picture is true. The explainer is explicit that the information alone cannot tell you whether the digital content is true, accurate or factual, and that Content Credentials do not provide value judgments. A signed manifest on a staged image is a correct record of a staged image.
The reverse also holds. A file without a manifest is not proof of anything. Most photographs in circulation never had one.
C2PA and the labels on social networks
Meta announced in February 2024 that it would label AI-generated images on Facebook, Instagram and Threads using two open standards, C2PA and IPTC. It later changed the wording of the label from Made with AI to AI info, after photographers complained about images labelled for light AI-assisted edits. If your photo carries an AI marker because an editing tool wrote one, the cause is a metadata field, which you can inspect. The guide on why Instagram labels photos as AI covers the fields in detail.
What it means for PassReal
PassReal removes C2PA manifests, IPTC fields, XMP blocks and text chunks by re-encoding the image from a pixel array, then shows a report of how many markers were found and how many remain. On a Gemini export, 174 went to 0. That is a statement about declared provenance only. It is not a claim about detectors and not a claim about SynthID, which is a separate pixel watermark. The how-to for doing it by hand is how to remove C2PA from an image.
Questions
Is C2PA the same as Content Credentials?
Yes in practice. The specification calls the record a C2PA Manifest, and Content Credentials is the name used for the feature that shows it to people.
Does C2PA prove an image is real?
No. The C2PA explainer says provenance information alone cannot tell you whether content is true, accurate or factual. It tells you who signed the claim and whether the content changed after.
Can a C2PA manifest be removed?
Yes. The explainer answers yes when asked, and tools such as ExifTool can delete the JUMBF block from several formats. Soft bindings are the approach for finding a removed manifest again.
Is C2PA the same as SynthID?
No. C2PA is metadata that sits beside the pixels. SynthID is a watermark written into the pixels of media made with Google tools. A file can have either, both or neither.
Which platforms read C2PA?
Meta said in 2024 that it reads C2PA and IPTC signals to label AI images on Facebook, Instagram and Threads. Support elsewhere varies, so check each platform's own documentation.
Limits
Naturalization operates on the pixels and on the file, not on the content of the image. A subject that is implausible - six fingers, inconsistent reflections, text that does not read - stays implausible after processing, and a human reviewer will notice it. PassReal changes what a statistical classifier measures, not what a person sees.
Results vary by generator, by subject and by detector, and detectors are retrained. No pass rate is published on this page because none has been measured in a way that would still hold next month.
Three images, no card, no expiry.