If you have opened an image lately and seen a small "Content Credentials" badge, or an "AI info" label in a search result, you have met C2PA. It is the technical standard behind most of the content provenance labelling rolling out across cameras, editing tools, and AI generators in 2026. This guide explains what a Content Credential actually contains, how it is attached to a file, why it can be removed by something as simple as a screenshot, and how our image metadata remover handles it.
What C2PA is
C2PA stands for the Coalition for Content Provenance and Authenticity, a standards body backed by Adobe, Microsoft, the BBC, camera makers, and others. The consumer facing brand for the signals it produces is Content Credentials. The idea is not to detect fakes or to hide a watermark in the pixels. It is to attach a tamper evident record of how a piece of media was made and edited, cryptographically signed so a viewer can check it has not been altered since signing. You can read the full technical specification at spec.c2pa.org.
What is inside a Content Credential
A C2PA record is called a manifest, and it is embedded in the file's metadata. A manifest is built from three kinds of thing:
- Assertions. Individual statements about the asset: the capture device, a creation or edit timestamp, the software used, whether generative AI was involved, and optionally the creator's identity. "Made with an AI tool" is just one possible assertion.
- Claims. A claim gathers the assertions together and references the exact bytes they describe, so the set cannot be quietly reshuffled.
- A signature. The claim is signed with a certificate from a trust list. This is what makes the credential tamper evident: change the pixels or the assertions and the signature no longer validates.
Chained together across edits, these manifests form a provenance history: captured on this camera, opened in this editor, exported on this date. The current specification sits in the 2.x series, with associated ISO reference work, and the format keeps evolving.
Hard binding versus soft binding
The most useful thing to understand about C2PA is how the credential is tied to the actual content, because it explains both the strengths and the gaps. There are two mechanisms.
- Hard binding. A cryptographic hash, typically SHA-256, is computed over the file's bytes and stored in the manifest. Any change to those bytes breaks the hash, so hard binding gives strong tamper evidence. Its weakness is fragility: ordinary operations like re-compression or a format change also break it.
- Soft binding. A perceptual hash or an invisible pixel watermark is used so the credential can be re-associated with the content even after minor edits or transcoding. Soft binding provides durability, at the cost of weaker security guarantees than a byte exact hash.
So called durable Content Credentials combine the two, pairing signed metadata with a soft binding so that if the metadata is stripped, a matching record can sometimes still be found in a cloud database. That is the industry's answer to the obvious problem, which we come to next.
Why a screenshot removes it
Here is the honest limitation. In the common case, a Content Credential lives in the file's metadata block, not in the pixels. That means it is removable. Re-saving the image in most editors, converting the format, uploading to a platform that rewrites files, or simply taking a screenshot produces a new file with no manifest attached. Nothing about the visible image changes. The provenance record just is not carried along. Soft bindings and cloud lookups can partially recover this, but only where those systems are deployed and consulted, which is far from universal today.
This is not a flaw we invented to sell a tool. It is inherent to metadata based provenance, and the C2PA community is open about it. It is also why regulators who lean on this technology, such as the drafters of the EU AI Act transparency rules, ask for content to be marked "as far as this is technically feasible."
C2PA is not a pixel watermark
A common misconception is that Content Credentials hide a mark inside the image itself, the way an invisible pixel watermark does. In the usual case they do not. C2PA is signed metadata that travels alongside the pixels, which is precisely why it is both easy to read and easy to lose. This is a different design philosophy from pixel watermarking, and the two are often confused because both get called watermarks in casual writing. Pixel watermarking embeds a signal in the image data that can survive re-saving but is hard to inspect and can degrade quality. Metadata provenance keeps the pixels untouched and stores a rich, human readable history, at the cost of being strippable. Understanding that split explains why the industry increasingly pairs them: metadata for the detailed provenance record, a soft binding for durability when the metadata is removed.
What assertions can actually tell you
Not every credential carries the same weight, and reading one well means checking what it actually asserts and who vouched for it. A credential that says only "exported by an editing app" tells you little about whether AI was involved. One that includes a specific "generated by AI" assertion, signed by a recognised provider, is far more informative. The signer matters too: a signature is only as trustworthy as the certificate behind it and the trust list it chains to. When you inspect a credential, the useful questions are what is being claimed, who signed it, and whether the signature still validates against the current bytes. A green badge alone is not the whole story.
How to inspect a Content Credential
You do not need our tools to look. The official Content Credentials Verify page lets you drop in an image and see any manifest it carries: the assertions, the signer, and whether the signature validates. If you get "no Content Credentials found," the file either never had one or has had it stripped by a previous save. Camera makers are starting to sign at capture, Samsung's Galaxy S25 line, for instance, can write C2PA data directly from the native camera, which is what is pushing the standard from a professional niche toward the mass market.
How our image tool handles C2PA
Our image metadata remover treats C2PA the same way it treats EXIF, XMP, and IPTC: it removes the metadata block and hands you back a clean file, without touching a single pixel. That matters for two honest reasons. First, removing metadata is legitimate privacy hygiene, EXIF alone can leak GPS coordinates, device serial numbers, and timestamps you may not want public. Second, we do not pretend to alter the image itself, because we do not. If a file carries a soft binding or is registered in a durable credentials database, stripping the local manifest does not guarantee the provenance is gone from every system, and we say so rather than promising a clean break we cannot deliver.
Should you remove it?
That depends entirely on what you do next, not on the act of removal. Clearing metadata from a photo you own before posting is routine and reasonable. Stripping a credential specifically to pass AI generated media off as something it is not, in a context where a viewer reasonably expects to know, defeats the purpose the standard exists to serve. The tool is neutral; the representation you attach to the result is not. We built the remover to do the mechanical job well and to be clear about where its power ends.
A short history, and where it is going
C2PA grew out of two earlier efforts, Adobe's Content Authenticity Initiative and Project Origin, which merged their approaches into a single open specification. The early versions were aimed at professionals: newsrooms proving a photo was unaltered, creators attaching authorship. The 2.x series tightened the security and reliability of the format, and adoption has since spread from that professional niche toward everyday devices, with camera makers signing at the point of capture and major platforms beginning to display credentials in their interfaces.
The frontier now is durability. Because a plain manifest is so easy to strip, the ecosystem is building durable Content Credentials that pair the signed metadata with a soft binding and a cloud lookup, so a stripped image can sometimes be matched back to its original record. This is promising, but it is not magic, and it is not universal. It only works where the soft binding survives the edit and where the checking system actually queries the database. For now, the honest summary stands: when a credential is present it is trustworthy, and when it is absent you have learned very little.
Frequently asked questions
What is C2PA in simple terms? It is a standard for attaching a signed, tamper evident record of how a piece of media was made and edited, including whether AI was involved. The consumer brand for it is Content Credentials.
Does C2PA prove an image is real? It proves a credential is present and unaltered since signing, by a particular signer. It does not prove an image is human made, and its absence proves nothing, because credentials are easily stripped.
Can Content Credentials be removed? Yes. In the common case they live in metadata, so re-saving, converting, or screenshotting a file drops them. Durable credentials add a soft binding and cloud lookup to make this harder, but not impossible.
Does removing C2PA change the image? No. Our tool removes the metadata block without touching a single pixel.
Related: Remove image metadata and EXIF · How our removal works · About this tool