A match compares an image with an enrolled identity
Face recognition compares information from a camera frame with the reference created during enrollment. Lunlo performs that processing locally. Its development policy asks for three consecutive matching frames, and resets the evidence when the identity changes, a face disappears or multiple faces appear.
Those consistency rules describe how a decision is assembled. They do not provide a measured false-acceptance or false-rejection rate. A displayed match should be interpreted as the model's current result, not as a completed unlock or a proven identity guarantee.
Liveness checks examine additional evidence
The liveness stage is intended to evaluate evidence beyond whether an image resembles an enrolled face. The current implementation requires liveness checks to pass. If its dedicated recognition model is unavailable, it stops instead of substituting a generic image comparison.
These software checks do not establish that a regular RGB camera can reliably distinguish a person from every printed image or replayed recording. The preview has not completed that validation. A photo or video may fool a system even when its ordinary matching demonstrations look convincing.
The camera and environment are part of the evaluation
Lunlo uses ordinary RGB camera input. It is not Apple's Face ID and does not provide Face ID depth sensing. The camera view, lighting, position and appearance during enrollment and later scans need to be evaluated together on the intended Mac.
The Face Lab view is designed to make recognition and liveness observations visible during development. Use it to identify which stage produces a result and when that result changes. A stable result in one room or camera arrangement does not establish performance in other conditions.
Real-world validation is still required
Local policy tests, model-loading checks and synthetic inference exercises can catch implementation errors. They do not replace tests of live enrollment, false acceptance, false rejection, printed-photo attempts, replayed video and real lock-screen behavior. Those outcomes remain unverified for the current preview.
Face recognition also does not replace the rest of the unlock sequence. The app still needs verified account access, an authorized local vault session and permission to type its locally encrypted password. Read the complete workflow and session requirements. The public service and download remain closed; consult release status before relying on availability.