Does this upload my photos anywhere?
No. FaceVersusFace runs entirely in your browser using WebAssembly
and TensorFlow.js. Your images are processed locally and never
transmitted to any server. Close the tab and everything is gone.
What technology powers the face detection?
face-api.js wraps TensorFlow.js models trained
for face detection (SSD MobileNet v1), 68-point facial landmark
detection, age & gender estimation, and expression
recognition. All models run client-side via WebAssembly.
What does the similarity score actually measure?
The score is a weighted composite of geometric ratios from 68
facial landmarks: eye aspect ratio, face width-to-height, jaw
width, nose bridge length, inter-eye distance, mouth width,
eyebrow arch, and face symmetry. It does not use deep face
embeddings — it is purely geometric.
Why does the page need to download models first?
face-api.js ships pre-trained neural network weights (~6 MB total)
fetched once from a CDN on first load. After that, all inference
runs locally. Weights are cached by your browser so subsequent
visits are instant.
What photos work best?
Front-facing, evenly lit photos with the face clearly visible
produce the most accurate results. Avoid extreme angles, heavy
filters, dark lighting, or sunglasses. JPEG, PNG, WebP, and GIF
are all supported.
Is this an identity verification system?
Absolutely not. FaceVersusFace is an entertainment tool based on
geometric ratios. It cannot reliably identify, match, or verify
anyone's identity. Do not use it for any decision-making related
to access, security, or personal rights.
Does it work on mobile?
Yes. The interface is fully responsive and the camera capture uses
the native device camera. Inference may be slower on older devices
but will work.
Do I need an API key or account?
No. FaceVersusFace uses only open-source software (face-api.js,
TensorFlow.js) and CDN-hosted model weights. There are no API
keys, no subscriptions, and no cost whatsoever.