What you get per face
- Filter probability and verdict band (High ≥ 0.52, Low 0.18–0.52, Natural < 0.18).
- AI-generation / face-swap assessment.
- Region reshaping ratings: jaw, nose, eyes, brows.
- Uncertain verdicts with the rule that triggered them, so reviewers know why.
How to use it responsibly
TrueFace output is supporting evidence. Under NIST SP 800-63A-4 and FATF customer due diligence guidance, the identity-proofing decision stays with your process. Route High and Uncertain results to manual review rather than auto-rejecting.
Error profile
On our preliminary N = 70 labelled set, a flag was correct 94.87% of the time but a “natural” result only 64.52%. Treat “natural” as absence of evidence, not proof.
Integration details on the API page; methodology in the whitepaper.