In short
- Active liveness asks for a gesture and costs conversion; passive liveness is invisible but puts more weight on the model.
- The attack class matters: printed photos, screen replays, masks and deepfakes are each caught by a different signal.
- In practice a hybrid flow wins: passive by default, active only when the score looks weak.
A face recognition model can be flawless and still worthless if what stands in front of the camera is a photo on a phone screen rather than a living person. That makes liveness, not recognition, the most fragile part of a biometric flow.
Two approaches, two different costs
Active liveness asks the user for a specific action: turn your head, blink, follow a dot on screen. Passive liveness asks for nothing — it analyses frame texture, light distribution, depth cues and micro-movement.
- Active: harder for an attacker, but every extra step costs 4–9% of users during enrolment.
- Passive: completes in 1.2 seconds on average and stays completely invisible to the user.
- Hybrid: passive runs first, and the active step is requested only when the confidence score is low.
99.6%
Liveness accuracy
1.2s
Average check time
4
Attack classes covered
40M+
Attempts verified
Which attack hits which barrier
It helps to split presentation attacks into four classes. Each class is caught by a different signal, which is exactly why no single model closes all of them.
- Printed photo: paper texture and the way light reflects are immediately visible to a passive model.
- Screen replay: moire patterns, frame rate and display brightness give it away.
- Silicone mask: it is the absence of micro-expression, not a depth map, that reports the problem.
- Deepfake stream: this needs camera-level attestation plus temporal consistency analysis.
The liveness question is not «is this person alive?». It is this: did the signal from the camera come from a real person, in real time?
What to choose for your integration
If your flow is mass enrolment, start passive — conversion decides here. If the flow carries real risk, such as a large transfer or signing a document, making the active step mandatory pays for itself.