How do Face Liveness Detection and Recapture Detection work?
Face Liveness Detection and Image Recapture Detection help you catch spoofing attempts, where a user submits a reproduction of an existing image instead of a genuine capture. They are commonly used in identity verification, KYC and fraud prevention workflows.
Face Liveness Detection
The Face Liveness Detection model determines whether a face is physically present in front of the camera at the moment of capture, or if the image is a presentation attack. It performs passive liveness detection: it works on a single photo or frame and requires no interaction from the user.
The model detects a wide range of attacks: faces displayed on screens, printed photos, faces added in post-processing, 2D and 3D masks, cutouts, deepfakes and AI-generated faces.
Before assessing liveness, the API checks that the image is suitable: a single main face must be present, the face must cover between 1% and 95% of the image, and it must be clearly visible. Images that don't meet these criteria are rejected with a rejection reason, and no liveness score is computed.
The API returns a score between 0 and 1. The closer the score is to 0, the more certain the model is that the image is a spoofing attempt.
Image Recapture Detection
The Recapture Detection model determines whether a photo was taken of an already-existing image, such as a picture displayed on a monitor, a printed photo, a magazine page or a poster. Recapture attacks are commonly used to bypass originality checks, strip watermarks or impersonate legitimate content.
Screenshots are digital captures and are classified as originals. Scanned documents are not considered recaptures either.
The API returns a score between 0 and 1. Scores above 0.5 generally indicate a recapture. For identity verification workflows, you can use a higher threshold such as 0.8.
Recapture Detection does not tell whether the original content is real or synthetic. It can be combined in the same request with AI-generated Image Detection and Deepfake Detection.
Operations and availability
Each image analyzed with Face Liveness Detection counts as 10 operations, and each image analyzed with Recapture Detection counts as 3 operations. Both models are available on paid plans.