Attractiveness Test

Entertainment is the honest label, and it follows from what the tool can reach. A face becomes coordinates, coordinates become ratios, ratios get compared against reference figures. Nothing in that chain touches how a person moves, speaks, laughs or is known to the people around them.

So the output ranks proximity to a mean, not people. Treating a high or low figure as a verdict misreads what was computed. The components are worth seeing separately.

Geometry splits into three ideas with three levels of support behind them.

Averageness

In 1990 Langlois and Roggman published work in which composite faces, many portraits digitally averaged together, were rated more attractive than most of the individual faces feeding the composite. The effect grew stronger as more faces were added. Averageness here is mathematical: close to the population mean on each measurement, not plain.

Explanations differ. Some read it as a signal of genetic heterozygosity. Others point to processing fluency, the finding that stimuli near a familiar prototype are easier for the visual system to handle and get rated more positively for that reason alone. So the effect may describe perception more than faces.

Symmetry

Rhodes and colleagues ran a long series of studies on facial symmetry and rated attractiveness, generally finding a positive but modest association. It runs smaller than the averageness effect and swings with how symmetry gets manipulated. Perfectly mirrored faces often strike observers as wrong.

Every real face is asymmetric. Chewing side, sleeping position, dental history and developmental noise leave measurable left-right differences in everyone. A symmetry figure describes a face. It does not rank one.

Neoclassical and phi canons

Renaissance drawing conventions gave painters teaching heuristics: the face divides into equal vertical thirds, spans five eye-widths, and carries a mouth one and a half times the nose width. Phi scoring checks whether facial distances approximate 1.618.

Tested against real populations, these canons largely fail. Study after study finds most people, including people rated highly attractive, miss the neoclassical proportions. They describe a studio convention, not a distribution of human faces.

Whichever component, the photograph decides the arithmetic.

Consistency, not improvement, is the goal. Anything changing the geometry in the image changes the number.

• Use even, diffuse front lighting. Side light deepens one cheek, and the algorithm reads that shadow edge as a jaw contour.
• Face the lens straight on with your eyes level with it. Looking up or down shifts the apparent vertical thirds by a wide margin.
• Hold a neutral expression, mouth closed. A smile widens the mouth, lifts the cheeks and narrows the eye aperture.
• Pull hair clear of the forehead, temples and jaw so the outline is visible.
• Stand back and crop in. At arm's length, lens perspective enlarges the nose against the ears.
• Remove glasses and shoot against a plain background.

Even a perfect photograph cannot rescue the premise.

Judgments about faces vary enormously across cultures and across time. Preferred face shape, weight, skin tone, feature size and grooming conventions differ between societies and shift inside one society across decades. Any fixed geometric standard captures one narrow set of assumptions.

Attraction as people experience it is also not a still image. Movement, voice, expression, warmth, familiarity and context all carry it. Research uses static photographs because the other variables resist control, a convenience of method rather than a finding that they matter less.

And the measured effects are population-level tendencies. A tendency across thousands of ratings tells you nothing dependable about one face. Individual variation swamps the effect size.

Landmark placement carries error of its own. MediaPipe holds to a few pixels under good conditions and degrades with poor light, heavy makeup, motion blur, occlusion or unusual pose. Two photographs of one person minutes apart routinely differ by several points, which shows what the number is worth.

Reference distributions are limited by whatever sample built them, and facial measurement datasets carry well-documented demographic skew. A score computed against a narrow population measures conformity to it, nothing more.

Nothing here is a medical, psychological or dermatological assessment, and no result should inform any decision about a cosmetic procedure. If you find yourself rechecking a score, or feeling worse after using tools like this one, close the page. Distress about appearance is common and treatable, and it is not something a website number can help with.

Your photo never leaves your device. Processing runs locally, no image reaches a server, and nothing remains once the page closes.

An attractiveness test here computes geometric proximity to a statistical average and prints it for entertainment, never as a reading on beauty, worth or desirability. Averageness carries the strongest evidence and stays modest, symmetry carries less, and the neoclassical and phi canons fit real faces badly. Even light and pose move the figure by several points, and cultural standards shift the target entirely. So take the number as arithmetic on distances. Run the test above if you are curious, then look at the face symmetry test or the golden ratio calculator for the individual measurements behind it.