Liveness that runs once and never asks "is this really happening right now" is a face-match step, not a liveness check. The two get sold as one thing.
Not "can you detect a face". It is can it tell a live person from a replay, a mask or a printed photo, and can it do that without turning onboarding into a lab test.
Face matching against the document photo or the NFC chip photo, whichever is stronger.
Liveness, active or passive depending on the risk level, checking that the capture is happening now rather than being replayed.
Presentation attack detection, tested to iBeta level 2: printed photos, screens, masks.
iBeta level 2 presentation attack detection, across the same capture pipeline as photo ident (still image, video stream, burst). This page is the liveness and matching layer on top of it.
A false accept on liveness is a deepfake or a replayed video that passes as a live person. A false reject is a legitimate user rejected because of lighting, glasses or a low-end camera, and it is the one that happens far more often and gets noticed far less.
Active versus passive liveness, and the PAD threshold, set per market and per risk class.
A liveness check that only matches a face is a control you can be told worked when it did not.
The capture and the reason a case failed, not just a pass or fail.
Tested to iBeta level 2 by an independent lab, not self-attested. That is the difference between a liveness claim and a liveness result.