Biometric verification only works if the biometric input is genuine. A system that can perfectly match a face to a government ID photograph is useless if the face being presented is a photograph held up to the camera, a video replay on a tablet screen, or a 3D-printed mask crafted from the target's social media photos. Presentation attacks — attempts to fool a biometric system with a fake representation of the target's biometric — are the primary attack vector against facial verification, and liveness detection is the technology designed to stop them.

The simplest form of liveness detection is active: the system asks the user to perform a specific action — turn their head, blink, smile, or follow a moving object with their eyes. The assumption is that a photograph or pre-recorded video cannot respond to dynamic prompts. This approach was effective against basic attacks in early systems, but it has significant weaknesses. Pre-recorded videos can be produced that include the required actions. Deepfake face swaps can respond to prompts in real time. And the friction of asking users to perform actions reduces completion rates and degrades the user experience.

Passive liveness detection operates differently. Instead of asking the user to do something, it analyses the biometric capture itself for signals that distinguish a live human from a fake. These signals include texture analysis — real skin has micro-texture patterns that differ from the flat surface of a printed photograph or screen display. They include depth analysis — using the slight perspective differences captured during a natural micro-movement to distinguish a three-dimensional face from a two-dimensional image. They include light reflection patterns — real skin reflects light differently from paper, plastic, or screen glass.

The advantage of passive liveness is that the user does not need to do anything beyond looking at their phone camera, which is what they were going to do anyway. The analysis happens in the background, adding no perceptible friction to the verification flow. The user takes a selfie. The system simultaneously performs biometric matching against their identity document and liveness analysis to confirm the selfie is from a live, present human. The result is returned in seconds.

The challenge is that the attack landscape has evolved beyond photographs and videos. Real-time deepfake face swaps — where an attacker uses software to overlay the target's face on their own in a live video feed — represent a fundamentally different threat. A deepfake face swap is not a static presentation attack. It moves naturally, responds to prompts, and maintains consistent depth characteristics because the underlying face is real. Only the appearance has been changed.

Detecting deepfake face swaps requires analysis that goes beyond traditional liveness detection. It requires examining the video feed for the computational artefacts that deepfake generation produces — subtle inconsistencies in skin texture, lighting interaction, temporal coherence, and edge blending that are not visible to the human eye but are detectable by trained machine learning models. This is the domain of deepfake detection, which operates as a complementary layer alongside traditional liveness detection.

The combination of passive liveness detection and deepfake screening represents the current state of the art in presentation attack prevention. It defeats photographs, video replays, 3D masks, and real-time deepfakes in a single, frictionless capture.

deepidv's identity verification platform includes both passive liveness detection and deepfake screening as standard components of every biometric verification, ensuring comprehensive protection against the full spectrum of presentation attacks.