A face recognizer never stores your photo. It stores a vector of ~512 numbers, your faceprint. Drop in a face, watch a generic one get rebuilt from that vector alone, then grab the numbers and scrub them by hand. The lesson the page makes playable: a faceprint is not anonymous. You can walk it back into a face.
the photo never leaves your device: alignment, embedding and decoding all run in this tabYour photo is aligned and run through ArcFace, a real face recognizer, which outputs a 512-number embedding: the same kind of vector that face-recognition systems store and match against. A small decoder (trained here) reads only that vector and rebuilds a generic face that lands on almost the same embedding. It never sees your pixels, only the faceprint.
The similarity meter re-runs the recognizer on the rebuilt face and compares its faceprint to yours. When it sits above a recognizer's ≈0.35 “same person” line, the rebuilt stranger would be matched as you. So the 512 numbers people call “anonymous” carry your identity, reversibly.
This is the inverse of recognition: it matches you against no database and names no one. It only ever touches the face you bring, entirely on your device.
Everything, face detection, alignment, the recognizer, the decoder, runs in this browser tab via ONNX Runtime Web. No photo, no embedding, and no pixel is ever uploaded. Reload and it's gone.