Use Case
Every AI output. Machine-labeled at the moment of creation.
C2PA manifests + invisible watermarks for AI-generated content.
AI content labeling means attaching a machine-readable record to every AI-generated image, audio clip, or video at the moment of creation. Verbitas does this in two layers: a C2PA 2.4 manifest (embedded in the file, readable by any compliant verifier) and an invisible watermark (embedded in the pixels or audio signal, survives metadata stripping).
How it works
- 01 Call POST /v1/sign with the AI-generated file and the image-genai-v1 recipe
- 02 Verbitas embeds a C2PA 2.4 manifest recording the AI tool, timestamp, and signer
- 03 TrustMark (image) or AudioSeal (audio) watermark is embedded in the signal
- 04 Manifest digest is indexed for soft-binding recovery
- 05 Any downstream verifier can check the manifest or detect the watermark
Technical notes
The C2PA manifest is readable by Adobe Photoshop, Leica cameras, Nikon cameras, and any C2PA-compliant reader. The TrustMark watermark survives JPEG recompression to 50% quality, resize to 25%, and 30% crop.
Limitations
AI content labeling establishes the provenance record at signing time. It does not retroactively label content that was generated before integration. It does not prevent an actor from generating unlabeled content using a non-integrated tool.