How KYC Platforms Speed Up Onboarding With a Face Detection API

[navigation]

Key takeaways

• Banuba's face detection API reads more than ten facial parameters on-device, including heart rate, skin tone, facial expressions, head position, and distance from the camera, so a KYC app can reject a bad capture before it is ever submitted.
• Banuba stores and transfers no personal data: face tracking and analysis run locally on the user's device, which is what makes the capture step defensible under GDPR.
• Detection is the front half of the flow. Banuba's biometric match and active liveness triggers handle the verification half, from the same SDK, so it is one integration rather than two.

Why does the capture step decide onboarding speed?

Every retry is a chance for the applicant to give up. A face that is half out of frame, turned too far, lit from behind, or covered by a scarf produces a submission the verification step rejects, and the applicant is sent back with no explanation of what went wrong.

A face detection API moves that judgment into the camera view. Instead of validating after upload, the app validates continuously while the camera is open: it knows where the face is, whether one face or several are present, and whether conditions are good enough to proceed. The applicant is corrected in the moment rather than rejected minutes later.

Banuba's face module is built for those conditions. It tracks up to 68 points on a face mesh with 37 morphs, holds tracking in low light and at distance, and stays reliable when the face is partly covered, including by a medical mask. Multi-face detection matters here too: if a second person is in frame, the app can say so before the capture is accepted.

How does Banuba's face detection API support a KYC flow?

Banuba's face detection API surfaces more than ten detectable facial parameters in real time, among them heart rate, skin tone, facial expressions, head position and tilt, distance from the camera, eyewear and facial hair detection, face shape and pupillary distance. For onboarding, the useful subset is small: is a face present, is it the only face, is it square to the camera, is it close enough, is anything covering it.

The second capability is the one compliance teams ask about first. Banuba performs no storing or transferring of personal data, because detection and analysis run on-device rather than in a cloud service. The frames used to judge capture quality never leave the phone, which shrinks the biometric data the platform has to account for. Weigh that against a wider comparison of on-device and cloud-based detection approaches before committing to an architecture.

Verification is a separate job from detection. Banuba's SDK also provides biometric match, comparing the face in the frame against a single reference image, and active liveness triggers such as blink, head movement, and prompted actions. Capture quality, liveness challenge, and 1:1 match therefore run through one integration instead of a tracking library stitched to a separate identity vendor.

What does the integration take?

Banuba Face AR SDK runs on iOS 13+ and Android 8.0+ with a 1280x720 camera recommended, at a minimum of 30 FPS, and supports Web, Windows, macOS, Unity, Flutter and React Native from the same codebase. That range matters when applicants arrive on a browser, an Android phone, and an iPhone in roughly equal numbers.

Teams building an iOS face detection API integration start from the Banuba iOS quickstart and the iOS samples repository, with the Face AR SDK documentation covering the tracking and feature configuration a capture screen needs. The face landmarks guide covers reading point coordinates directly, which is what a custom capture-quality check is built on.

Banuba has no published customer case for a KYC deployment, so treat the figures above as product specification, not a reported outcome from a named platform.

Start with a trial

Test the capture step against your own applicants before you change anything downstream. Request a 14-day trial token for the Banuba Face API and run it on the devices your users actually onboard from.