Calabi Labs ? SEO guide ? 2026-07-19
How Sora C2PA metadata works, why cropping is not enough, and how to strip AI provenance before posting.
Sora files can carry C2PA Content Credentials and AI-origin metadata that survives normal editing. To clean the file, you need to remove the C2PA/JUMBF container, XMP AI flags, generator identifiers, and encoder history, then verify the output before upload. A crop only addresses visible marks; it does not clean the provenance layer.
C2PA is a provenance standard. In plain English, it lets software attach a signed history to media. That history can say the file was created or edited by an AI system. Platforms, publishers, and verification tools can inspect that history without asking you what tool you used.
For creators, the practical issue is simple: if the file still contains the AI-origin manifest, the platform does not need a complicated visual detector. The label can come from metadata alone.
Calabi strips the metadata/container layer, re-encodes the file through a clean pipeline, injects a phone-capture profile, then returns a forensic proof card. The proof card is the key. You do not have to guess whether the C2PA references are still there.
For a bottom-of-funnel workflow, start with Remove Sora watermark, then pick a device identity such as iPhone 17 Pro metadata.
The common mistake is using a visual watermark remover and assuming the file is clean. If the C2PA manifest and AI XMP flags are still present, the file is not clean at the layer platforms can scan fastest.