C2PA (Coalition for Content Provenance and Authenticity) manifests — the signed records that AI tools like DALL·E and Adobe Firefly embed to disclose how an image was made. This is the exact signal Instagram, Facebook, and TikTok read to apply "AI info" / "AI-generated" labels. AiEraser doesn't just detect that a manifest exists — it decodes the actual structure: claim generator, recorded actions, and signature algorithm.
caBX, directly after IHDR.C2PA.In every case, the payload is a JUMBF (JPEG Universal Metadata Box Format) container holding CBOR-encoded claims — not a simple text field, which is why many basic "metadata remover" tools miss it entirely if they only check for standard EXIF/XMP tags.
Based on real manifests we've decoded: a claim generator name (identifying the producing service), a chain of recorded actions (created, converted, sometimes watermarked), a digital-source-type flag, and a cryptographic signature (commonly ES256) tying the manifest to the issuer's certificate.
No. C2PA is one provenance signal, not a universal detector — removing the manifest just means that particular signal is no longer present. It doesn't affect pixel-level watermarking or visual AI-classifiers some platforms also run.
AiEraser's View mode gives you the full decoded breakdown — claim generator, every action, signature algorithm — on your own file, and Remove mode strips it entirely.