When a client asks whether an image is original, “look at the metadata” is too vague to be a useful answer. Metadata can be edited or removed, and a file can be copied without its history. C2PA Content Credentials address a different problem: recording a signed, inspectable history of an asset and the actions performed on it.
The distinction matters to AI creators, photographers, agencies, and small businesses. You may want to remove private fields from a public copy while keeping a provenance record for an internal archive. You may also want to know why a clean re-encode does not automatically preserve a credential. Unmetaimage is a local image re-encoding tool; it does not issue or validate C2PA credentials.
What C2PA and Content Credentials mean
C2PA is an open technical standard for content provenance and authenticity. In the C2PA Content Credentials specification, a Content Credential is the non-technical name for a C2PA Manifest. A manifest contains a claim, assertions about the asset or actions, and a claim signature. A manifest store can be embedded in an asset or stored separately and referenced.
“Authenticity” in this context has a precise meaning. The specification describes it as facts about digital content that can be cryptographically verified as not having been tampered with. That is evidence about the relationship between a file and a signed record. It is not a universal judgment that the image is true, fair, or good, and it does not tell a viewer whether the scene itself happened in the physical world.
The parts you will see in a credential
An assertion is a statement about an asset, such as an action, an ingredient, a capture detail, or metadata collected when a claim is created. A claim references those assertions and is digitally signed. The signer’s credential is checked against the validator’s trust model. The manifest also includes a content binding that associates the record with the digital content.
The binding can be hard or soft. The specification defines a hard binding as one or more cryptographic hashes that identify all or part of an asset. A soft binding can use a fingerprint or an invisible watermark to help discover a manifest when the file and its manifest become separated. That is why “Content Credentials” should not be reduced to a visible badge or a single EXIF field.
The result is a chain of recorded changes. A later application can add a new claim describing a crop, color correction, or other transformation. A validator can then report whether the manifest is well-formed, whether its signature checks out, and whether the covered content matches. Those reports are useful trust signals, but they still require a human to understand who signed the claim and what the assertion actually says.
What creators should record
If your work may need to be audited or handed to a client, keep the generation and editing context in an archive even when you publish a small web copy. Useful records can include:
- the original file and its checksum;
- the application or pipeline that created it;
- prompt, seed, model, and edit actions when those are relevant;
- the version of a credentialed file and the signer identity;
- the exact derivative delivered to a client or platform.
Do not assume that every generator writes a C2PA manifest or that every platform displays one. Availability depends on the tools in the workflow. If a credential is important, inspect the file with a C2PA-aware validator and keep the validator’s report with the asset. A prompt saved in a PNG text chunk can help with reproducibility, but it is not the same thing as a signed provenance claim.
What happens when you clean or re-encode an image
An ordinary browser re-encode starts with decoded pixels and creates a new JPEG, PNG, or WebP file. It may drop embedded EXIF, XMP, prompt text, or a C2PA manifest because those records are not part of the pixels being drawn. The new file should therefore be treated as a derivative with a new identity. It does not inherit the original credential merely because it looks identical on screen.
The C2PA specification also describes how a manifest can be found through embedded locations or external references. A tool that removes file metadata can break those discovery paths, while a repository or a separate copy may still retain the old record. For that reason, preserve the credentialed original before using a cleaner. The image metadata guide covers the privacy side of making a sharing copy.
A practical decision for a small business
Use credentials when your value comes from a traceable creation and editing history: commissioned campaigns, editorial work, product imagery, or collaborations with multiple contributors. Keep the signed source and the delivered derivative together in your records.
Use a local metadata cleanup when your immediate goal is to reduce attached private information before sharing. Unmetaimage processes the selected image in the browser and leaves the original unchanged. Check the downloaded copy afterward, because no browser export can promise that every possible metadata field, external reference, or pixel-level signal has disappeared.
If your question is whether an image is AI-generated, combine provenance with context. A missing credential is not proof that an image was human-made, and a valid credential is not a claim that every visual statement is true. For the difference between file metadata, invisible watermarks, and detector scores, read metadata versus invisible watermarks and AI detection.