Durable Content Credentials: three layers
Manifest, invisible watermark and perceptual fingerprint survive different attacks and each fails somewhere. What your provenance claim actually proves.
Manifest, invisible watermark and perceptual fingerprint survive different attacks and each fails somewhere. What your provenance claim actually proves.
Meta's AI Info label fires from four independent signals. Which ones you control, which you don't, and how a mostly-photographic image ends up labelled.
TikTok watermarks uploads that arrive with C2PA credentials attached. That makes attaching provenance a labelling decision your toolchain is already making.
A C2PA manifest is a hash over exact bytes, so re-encodes and screenshots kill it by design. How to find the hop in your pipeline where provenance is lost.
A Content Credential is not a watermark — it is a signed manifest of assertions about origin and edits. What is actually inside one, and what survives editing.
A Content Credential is two different bindings doing two different jobs. Walk a file through trims, re-encodes and uploads and see which one actually survives.
SynthID, a platform AI label, and a signed manifest all get called watermarking. What SynthID embeds per modality, and what it cannot tell you.