CONTENT CREDENTIALS
C2PA Metadata: Content Credentials and Cleanup
A Remove AI Label guide to reading embedded provenance and deciding when a cleaned copy fits your delivery.
A record, not a verdict
Content Credentials can describe capture, editing and generation. A C2PA container alone does not prove AI origin. Authentication requires a credential verifier, not a metadata-presence report.
Read the report
Remove AI Label identifies supported JPEG APP11/JUMBF, PNG caBX and WebP provenance containers. A byte count describes the attachment, not who signed it or whether its assertions are trusted.
Clean only when appropriate
Inspect an original, clean a copy, then compare reports. Supported metadata is removed together, not selectively. Encoded pixels, color profiles and necessary orientation remain intact. Preserve a master when clients or archives need provenance.
Know what can remain
External manifests and platform records may survive. This browser-local workflow does not remove SynthID or guarantee platform AI-label changes. Use JPG, PNG or WebP; HEIC, AVIF, GIF, PDF, video and audio are unsupported. Removing metadata creates no new authorship or usage rights.
Distinguish presence from authentication
A report such as C2PA / JUMBF: 12 KB describes bytes in a supported container. It does not identify a trusted signer, validate a hash or determine which assertions remain accurate. Keep those questions separate when documenting a delivery. Download a cleaned copy, reopen it in the checker, and compare the output with the original. Small attachments may barely change file size. If the project requires an intact credential chain or selective editing, use the original application's export options instead of this complete metadata-cleanup workflow.
Further reading: C2PA specification · Google SynthID