Method, limits and privacy

How we measure

Every number on this site comes from fixed formulas applied to points on your face or samples of your color. This page publishes all of them.

Updated October 2026 · By the Hue & Shape team

Hue & Shape runs four free tools: a face shape detector, a symmetry test, a proportions reading and a color analysis with an undertone test. None of them is a black box. Each one turns a photo into a few measurements, then applies rules that are written down below, with the same constants the code uses. If a rule changes, this page changes with it, because its tables are generated from the code.

The measurements are geometry and color science, not judgments. A face shape is a label for three ratios. A symmetry index is an average distance. A color season is the nearest of twelve fixed points. Where a number is our own scale rather than a published standard, the section says so.

The landmark model

Every face reading starts with an open-source face landmark model, released under the Apache 2.0 license, that places 478 points on a face in a single photo. The model files and its runtime are served from this site and run in your browser with WebAssembly, so the image never leaves your device. The license notice ships with the site.

Each point has an x and y position as a fraction of the image width and height. Distances are corrected for the image aspect ratio, so a portrait photo and a square crop of the same face give the same ratios. The tools use a small set of the 478 points, listed here by what they sit on.

Landmark points used by the measurements
Measurement	Points used (model index)
Face length	Top of the face outline (10) to chin (152)
Forehead width	Left and right temple (127, 356)
Cheekbone width	Left and right cheekbone (234, 454)
Jaw width	Left and right jaw (172, 397)
Jaw angle	Angle at the chin (152) between the rays to the two jaw points
Scale	Pupil centers (468, 473)
Thirds	Top (10), brows (105, 334), nose base (2), chin (152)
Golden-ratio widths	Mouth corners (291, 61) and nose wings (129, 358)
Points in the full mesh	478

The top point is the highest point of the detected face outline at the center of the forehead. On most faces it sits at or just below the hairline; on a high or receding hairline it can sit lower than where the hair starts, which shortens the measured face length a little.

Face shape: three ratios and seven rules

The detector measures face length (top point to chin), forehead width (across the temples), cheekbone width and jaw width. It divides three of them by the cheekbone width: length to width, forehead to cheekbones and jaw to cheekbones. Cheekbone width is the reference because it sits between the hairline and the jaw, the two edges that hair and beards most often hide.

The rules run in a fixed order and the first one that matches names the shape. Order matters: a long face with a wide jaw is called oblong before the triangle rule is checked, because length is the more visible trait. These are the exact boundaries the code uses.

Face shape rules in the order they run
Order	Shape	Rule (L = length to cheekbones, F = forehead to cheekbones, J = jaw to cheekbones)
1	Oblong	L at least 1.60
2	Triangle	J at least 0.99, or J at least 0.95 with the forehead narrower than the jaw by 0.10 or more
3	Heart	F at least 1.00 and J at most 0.88
4	Diamond	F at most 0.90, J at most 0.90, and L at least 1.25
5	Square	J at least 0.94 and L at most 1.35
6	Round	L at most 1.20
7	Oval	Everything else: longer than wide with a gently narrower jaw

A face that lands near a boundary is common, so every result also reports a runner-up shape and a confidence. Confidence here is how far the ratio sits past the boundary that decided it, from 0.55 at the line to as much as 0.95 well clear of it. It is a statement about the measurement, not about the face.

These bands are this site's rules. Face shape names are a styling convention rather than an anatomical standard, and different guides draw the lines in slightly different places. Publishing ours means you can check a result with a tape measure and the same arithmetic.

Turning a photo into inches

A photo has no ruler in it, so any size in inches is an estimate. The tools scale the image by the distance between your pupils, set to the adult average below. The ratios that decide your shape do not depend on this step; only the inch figures in the full analysis do.

The scale reference
Reference	Value	Used for
Average adult pupil distance	63 millimeters (about 2.5 inches)	Inch figures in the full analysis and the frame width guide

Adult pupil distances vary by several millimeters either side of the average, so the inch figures carry roughly that proportion of error. They are useful for comparing your face with a frame chart, not as a fitting.

The symmetry index

Symmetry is measured on ten mirrored pairs of points. The face is mirrored about a vertical axis through the midpoint of the top point and the chin. For each pair, the tool takes the difference between the two points' distances from that axis and the difference in their heights, combines the two into one distance, and divides it by the distance between the pupils. That makes a large face in a close-up and a small face in a wide shot comparable.

The ten mirrored pairs
Pair	Points (left, right)
Eye corners, outer	263, 33
Eye corners, inner	362, 133
Brow ridge	105, 334
Cheekbones	234, 454
Temples	127, 356
Jaw, upper	172, 397
Jaw, lower	58, 288
Nose, alar	129, 358
Mouth corners	291, 61
Forehead, outer	127, 356

The index is 100 minus 500 times the mean pair difference, held between 0 and 100. In words: every 1% of pupil distance of average mismatch costs 5 points. A face whose pairs differ by 2% of the pupil distance on average reads 90; 4% reads 80. The scale is ours, chosen so ordinary photos spread across the upper part of the range. There is no published norm for it, so the tool does not tell you what a typical number is, and the number should only be compared with your own photos taken the same way.

Head turn moves this number more than anything about the face. A few degrees of yaw shifts every pair one way, which is why the quality check rejects turned heads before measuring.

Facial thirds and golden-ratio closeness

Facial thirds compares three heights: top point to the brow line, brow line to the base of the nose, and base of the nose to the chin. The tool reports each as a percentage of the total, then measures how far the three sit from an even split. The thirds index is 100 minus 260 times that spread (the mean distance from one third, as a fraction of a third). Thirds of 33, 33 and 34 percent score 97; thirds of 30, 33 and 37 score 81.

Golden-ratio closeness compares two ratios with 1.618: face length to cheekbone width, and mouth width to nose width. Each ratio scores 1 minus its distance from 1.618 as a fraction of 1.618, and the two are averaged. A length ratio of 1.40 and a mouth ratio of 1.50 average to 90. The proportion index on the attractiveness page is the mean of the thirds index and this closeness.

Both are descriptions of arrangement. The site does not claim the golden ratio defines beauty, and the attractiveness page reports these numbers without a score for a person. The upper third depends on where the top point sits, so a hairline that the model reads low will shrink it.

Photo checks before anything is measured

Before measuring, every face tool checks the photo and asks for a retake when a problem would move the numbers. It reports at most two problems at a time, in the order below, because the first fix often removes the second. These are the limits the code uses.

Photo quality limits
Check	Limit	What the tool asks you to change
One face	Exactly one face in the frame	Crop out or step away from anyone else
Too dark	Mean brightness below 0.30	Face a window or a soft light
Too bright	Mean brightness above 0.88	Move out of direct sun
Uneven light	Left and right halves differ by more than 0.13	Turn so the light falls evenly
Head tipped	Roll over 8 degrees	Level your eyes
Head turned	Yaw or pitch over 12 degrees	Look straight into the lens
Face too small	Face under 32% of the image height	Move closer or crop
Soft focus	Edge detail below 0.02	Hold still and refocus

Brightness is the mean relative luminance of the face region on a 0 to 1 scale. Head pose comes from the model's transformation matrix when the browser provides it; when it does not, the pose checks are skipped rather than guessed.

Color: undertone, depth, clarity and the twelve anchors

The color test asks you to tap three points: a cheek, an iris and your hair. Each sample is averaged from its middle so a highlight or a shadow does not decide it. The skin sample reads undertone from hue and saturation: a yellow to orange hue reads warm, a pink to red hue reads cool, and a muted, green-leaning sample (hue 35 to 100 degrees, saturation 0.10 to 0.36) reads olive. Hair and eyes add smaller votes.

Depth comes mostly from the hair: 62% from how dark the hair is and 38% from how dark the skin is. Clarity comes from contrast: 1.4 times the lightness gap between hair and skin plus 0.8 times the gap between eyes and skin. Seven quiz answers then add their own votes, because a phone camera's automatic white balance can push any skin sample warm or cool. The quiz can be taken without a photo.

The combined reading becomes three numbers: warmth from minus 1 (cool) to 1 (warm), depth from 0 to 1, and clarity from 0 to 1. The season is the nearest of the twelve fixed anchors below, with warmth weighted most. Neutral readings keep 60% of their lean before matching and olive readings keep 80% and shift slightly cool, because olive skin usually reads muted rather than golden.

The twelve season anchors
Season	Warmth (-1 cool to 1 warm)	Depth (0 light to 1 deep)	Clarity (0 soft to 1 bright)
Light Spring	0.55	0.18	0.62
Warm Spring	0.95	0.45	0.72
Bright Spring	0.50	0.50	1.00
Light Summer	-0.45	0.22	0.50
Cool Summer	-0.90	0.45	0.58
Soft Summer	-0.35	0.50	0.22
Soft Autumn	0.40	0.55	0.18
Warm Autumn	0.95	0.62	0.45
Deep Autumn	0.60	0.90	0.50
Deep Winter	-0.50	0.90	0.72
Cool Winter	-0.95	0.60	0.88
Bright Winter	-0.35	0.52	1.00

Distance weights: warmth 1.25, depth 1, clarity 0.8. Depth answers map to 0.2 (light), 0.5 (medium) and 0.82 (deep).

The anchors are this site's placement of the twelve seasons, not an industry standard; seasonal color systems disagree at the edges. They are drawn on every season page as a map, so you can see which seasons sit next to yours.

How the palettes are checked

Each season page lists 36 colors in six groups. Before a palette ships, an automated test converts every color to CIELAB, the color space designed so equal distances look roughly equally different, and checks the averages: the light seasons must have the highest mean lightness, the deep seasons the lowest, the bright seasons the highest mean chroma, each soft season the lowest chroma in its family, and every warm season a yellower average than every cool season. A palette written carelessly fails those checks.

Hex codes describe colors on a screen. Fabric, dye lots and print all shift them, so treat a hex value as a target to compare against, not a shopping code.

What the tools cannot see

Camera color is the largest limit for the color tools. Phones adjust white balance and saturation automatically, and a warm bulb makes almost everyone read warm. Daylight from a window in front of you is the most consistent light; the quiz exists because no light is perfect.

Camera distance changes face geometry. At arm's length, features closest to the lens look slightly larger than those further away, which narrows the face edges relative to the nose. A photo taken from a little further away and cropped gives steadier ratios than a close selfie.

Head pose, expression and hair matter for the face tools. A smile widens the mouth and lifts the cheeks, a chin tilt changes face length, and hair over the forehead hides the top of the face. Makeup does not move landmarks but does change a skin color sample.

None of the readings is medical, dermatological or cosmetic advice, and none rates a person. A measurement describes one photo.

Where your photo goes

Nowhere. The landmark model and the color sampling run in your browser, and the analysis makes no network request with the image. If you save a result to an account, the site stores the measurements (distances, ratios, the shape name and, for color, the three sampled color values), never the photo. The privacy policy lists everything that is stored and how to delete it.

Who writes this

This page and the guides across the site are written and maintained by the Hue & Shape team. When a formula, a boundary or a palette changes, the page that uses it shows a new updated date. Questions or corrections are welcome at hello@hueandshape.com, and a correction that checks out is applied to the page and the code together.

Sources

Dodgson, N. A. (2004). Variation and extrema of human interpupillary distance. Proceedings of SPIE 5291, Stereoscopic Displays and Virtual Reality Systems XI.