INPUT
One clear front view
Unusable images are rejected instead of receiving a guessed score.
FACE SYMMETRY TEST · FREE · ON-DEVICE
Symmetry is the heaviest of the four dimensions in the composite, and it is the one with the most explicit method. This face symmetry test mirrors 30 fixed landmark pairs across the measured midline, measures how far each mirrored point lands from its partner — sideways and vertically both — and maps the root-mean-square of those 30 misses to a 0–100 score.
Free scan · processed on your device · no photo upload · no account
INPUT
One clear front view
Unusable images are rejected instead of receiving a guessed score.
OUTPUT
0–100 and 0–10
Four integer sub-scores, measured values, and reference bands are shown.
PRIVACY
On-device only
Photos and measurements do not leave this browser for the free scan.
01
The facial symmetry check does not compare halves of your photo as images. It compares 30 pairs of detected landmark coordinates after the face has been rotated upright and rescaled. That matters: an image-level comparison is dominated by lighting and camera roll, while a coordinate comparison is not.
02
The pairs are fixed and published in the engine, five to each of six regions: brow, eye, cheek and zygomatic, nose, mouth, and jaw. The list is an ordered array rather than a set, because floating-point summation is not associative — reordering the pairs would change the last digits of the result, so the order is part of the formula.
03
No face returns zero. The reference band for the symmetry RMS runs up to 0.015 in face-width units, and the half-point constant is 0.030 — meaning a face whose RMS lands at 0.030 scores 50 on this dimension. Those two numbers are engineering constants, not population statistics. Nobody has measured where you sit among other people, and this page will not pretend otherwise.
04
Symmetry carries 35% of the composite, more than any other dimension. That is a design choice, not a finding: symmetry is the dimension least disturbed by the front-view projection, so it is the one the engine trusts most. Facial thirds and facial fifths carry 25% each, and the golden-ratio group carries 15%.
05
This is a front view, so the measurement runs in the image plane only. Depth asymmetry — one cheekbone sitting further forward, a deviated nasal septum in profile — produces no signal here. A turned head is rejected rather than scored, because a turned head manufactures asymmetry that is not in the face.
06
The symmetry test runs in your browser. The photo is decoded locally, the 478 landmarks are computed locally, and the 30 residuals are computed locally. No image, no coordinate, and no score record reaches our servers, and no account is created.
FAQ
It removes camera roll, rescales to face width, and mirrors 30 fixed landmark pairs across the measured midline. For each pair it measures how far the mirrored point misses its partner, counting both the sideways offset and the vertical one, then takes the root-mean-square across all 30. That single value is mapped to a 0–100 sub-score.
The engine publishes a reference band up to 0.015 RMS and a half-point at 0.030, so a smaller RMS means a higher sub-score. It does not publish a good or bad threshold, because it has no population distribution to draw one from.
The 30 pairs are anatomically identifiable on both sides — brow, eye, cheek, nose, mouth, jaw. Most of the 478 mesh vertices are interpolated surface points whose mirror partner is not a matching feature, so including them would add noise, not information.
Roll is removed before measurement, so a mild tilt does not. A tilt beyond the gate threshold is rejected outright and receives no score at all, rather than a score the tilt has quietly distorted.
Yes. There is no sign-up, email, or payment. The scan runs entirely in your browser and no photo or measurement is uploaded or stored.
Run the free scan in your browser. You will see the score, fixed weights, measured dimensions, and limits together.