Live demo
AdvancedTry it: face-gate lab
This is a face-similarity learning demo running in your browser, not a production admission system. A tiny face detector finds your face, the detected rectangle is cropped, and a recognition model compares it with one enrolled template. This prototype does not align landmarks, detect presentation attacks, or check event tickets. Enroll yourself, walk up to the gate, then run the stranger test.
Privacy: everything on this page runs on your device. The camera feed, the face templates, the comparisons — nothing is uploaded or saved by this demo. Images and templates remain in tab memory until cleared or the tab closes. This is not a guarantee of forensic memory erasure.
Step 1: load the models
Two small models download into your browser (about 7 MB total, served from this site): a face detector and a face-embedding model. This is the pre-trained model from the tutorial — you never train anything yourself.
Models not loaded.
Step 2: enroll
Sit facing your camera in even light, then capture. One face, one template — the same as the registration selfie in the tutorial.
Not enrolled.
Step 3: run the gate
The gate compares each arrival against your enrolled template and admits below the threshold. Try it yourself (admit), then have someone else step in front of the camera — the stranger test (deny).
Gate idle
distance: —
No-camera path: test with two photos
Upload one photo to enroll, then a second photo to test. Same face in both: admit. A different person: deny. This is also how you run the stranger test without a second person in the room.
Step 4: delete the template
Event over, templates die. This wipes the enrolled template from the page — the same deletion routine the tutorial puts on a checklist.
Gate log
Starter kit
The starter kit holds the consent and deletion policy template, the enrollment email copy, the gate setup checklist, the calibration and threshold worksheet, and pointers to the open-source embedding models this demo uses.
Want the files behind this lab, sample dataset included? Email ai@trihelix.co and we will send them your way.