Use Case
Not deepfake detection. Deepfake traceability.
Signed provenance records that make synthetic media traceable to its creator.
Verbitas does not claim to detect deepfakes. Deepfake detection is a separate, hard problem. What Verbitas provides is the infrastructure that makes deepfakes traceable: a signed C2PA manifest that records who created the synthetic content, what tool was used, and when. Combined with a disclosure assertion and an invisible watermark, this creates a verifiable record that the content is synthetic — and who is responsible for it.
How it works
- 01 Use the image-deepfake-v1 recipe to include a deepfake disclosure assertion in the manifest
- 02 The manifest records the creator, the synthesis tool, and a disclosure flag
- 03 TrustMark watermark is embedded for recovery after metadata stripping
- 04 Downstream verifiers receive the disclosure assertion in the manifest
- 05 The deployer is responsible for surfacing the disclosure to end users
Technical notes
The disclosure_assertion in the C2PA manifest uses the c2pa.ai_generative assertion type. Deployers must implement visible UI disclosure in addition to the machine-readable manifest. Verbitas provides the manifest; the deployer provides the user-facing disclosure label.
Limitations
Verbitas cannot retroactively mark synthetic content that was created without Verbitas. It cannot detect whether content was synthetically generated if no manifest is present. "Inconclusive" verification results mean no provenance record was found — they do not indicate synthetic origin.