Face Identity Matcher
Compare two face photos or videos using AI — runs entirely in your browser via ONNX Runtime Web. No upload, no server, 100% private.
How to use this face identity matcher
- Upload a reference photo (the known identity — clear face, good lighting)
- Upload a test photo or short video clip to compare against the reference
- Models are cached after the first load (~6 MB total)
- Wait for "Face embedded" badges on both cards, then click Compare Identities
- Read the verdict: GOOD ≥ 65% · RISKY 50–65% · IDENTITY MISMATCH < 50%
Related Tools
Frequently Asked Questions
Does it work on iPhone?
Yes — on iOS 17+ it runs via WebGPU (uses GPU memory, bypasses the WASM limit). On iOS 15/16 it uses sequential WASM loading (one model at a time) to cut peak memory in half. Both paths are tried automatically.
Are my photos sent to a server?
No. Detection and recognition run entirely in your browser. Your images never leave your device.
Which models are used?
SCRFD-500M (2.4 MB) for face detection and MobileFaceNet INT8 (3.4 MB) for recognition, from InsightFace's buffalo_sc pack.
How does the score work?
≥ 65% cosine similarity = same person (GOOD). 50–65% = uncertain (RISKY). < 50% = different person (IDENTITY MISMATCH).
Does video work?
Yes — upload a video as the test input. The tool samples 1 frame/second for the first 5 seconds and averages the embeddings.