The dials that carry a face and how far they travel

Which measurements make a face read male, female, older or from another population, ranked by how reliably they separate faces, with measured ranges by sex, a slider mapping that spends its codes where faces differ, and the data as JSON. Every figure attributed.

By Mackenzie Bowes · mackenzie@mackenziebowes.com · @_mackenziebowes on X

Head circumference by sex, adults.
Contents

The answer

Cheekbone width by sex (ANSUR II, n≈6,068).

A character creator needs a few dials that carry most of how a face reads. The measurement literature says which ones they are, and how far each should travel.

The obvious dials and where they break

Sex differences in facial width-to-height ratio are small and inconsistent.

Three shortcuts look right and fail.

fWHR as the masculinity dial. Facial width-to-height ratio is the most studied "masculine face" measure. Its sex difference is d = 0.11 across 10,853 people.[14] It vanished in athletes once body size was controlled,[15] it flipped sign in the army sample depending on how height was measured,[1] and after about age 48 women have the larger ratio.[16] Use cheekbone width instead.

The neoclassical canons as targets. Four of the eight classical proportion rules differ significantly between ethnic groups. The most variable measures (forehead height, eye spacing and nose width) are the canons' core assumptions.[13] Use them as a grid to deviate from.

A linear 8-bit slider. A byte gives 256 steps. Mapped linearly over ±3 SD, a third of the codes go to the 4.6% of people beyond ±2 SD, where neighbouring values look the same.

Take the dials and the data

The same 256 slider codes, spent two ways. A linear map gives a third of them to the 4.6% of people beyond ±2 SD; a quantile map gives every code an equal share of people.

Map each slider code through the inverse normal CDF, so every code is an equally likely quantile:

// code 0-255 -> a measure in mm or degrees; probit is the inverse normal CDF
const dial = (code: number, mean: number, sd: number) =>
  mean + sd * probit((code + 0.5) / 256)

Codes 0 and 255 land at about ±2.9 SD, so the reach matches a wide linear slider, with the resolution where faces actually differ. If the parameter can be a float, the constraint disappears.

The solid band is ±1 SD (about 68% of adults); the pale strip runs to ±2 SD (about 95%). Treat the edges as reach markers, not realism gates: going past them is stylisation.

DialFemaleMaleSource
Bizygomatic width#133.8 ± 5.6 mm142.6 ± 6.2 mmUS Army, n≈6,068[1]
Bigonial width#91 ± 5.9 mm97 ± 5.8 mmFarkas adult reference[2]
Head circumference#54.2 ± 2.4 cm56.3 ± 2.6 cmn≈2,508[9]
Interpupillary distance#61.4 mm63.9 mmTurkish adults, n=756[5]
Gonial angle#124.7 ± 3.2°122.2 ± 3.9°149 Iranian patients; direction disputed[3]
fWHR#1.921.98athletes, n=3,481; weak, see above[15]

The headline dials, by sex

These numbers as data, with means, SDs, samples, sources and a verification status per measure: parameters.json.

Every measure in full, by region:

What this says about how we work

The revision notes, one box each.

We take a craft problem to the measurement literature, check the headline numbers against their papers, cut every figure we can't cite, and ship the result as dials and data a character creator can load.

That's the reading we'd give your character pipeline before you build sliders on guesses.

If you're building a character creator and your faces keep drifting toward the same few looks, I'd like to hear about it: mackenzie@mackenziebowes.com or @_mackenziebowes.

Sources

  1. Caton & Dixson (2022). Bizygomatic width is highly sexually dimorphic when adjusting for allometry. Biology Letters. pmc.ncbi.nlm.nih.gov/articles/PMC9554718/
  2. Farkas adult craniofacial norms (gonial width M 97 ± 5.8 / F 91 ± 5.9 mm), as tabulated in Ogodescu et al. (2021), Updating Standards of Facial Growth. pmc.ncbi.nlm.nih.gov/articles/PMC8156684
  3. Gender Determination Using Mandibular Angle Measurement (mandibular angle M 122.19 ± 3.89 vs F 124.72 ± 3.24). brieflands.com/journals/amhsr/articles/114608
  4. Shaw & Kahn / Pessa: Changes in the Facial Skeleton With Aging. pmc.ncbi.nlm.nih.gov/articles/PMC3404279/
  5. Evaluation of interpupillary distance in the Turkish population (n≈756, by age/sex). pmc.ncbi.nlm.nih.gov/articles/PMC4529256/
  6. Quantitative Analysis of Male Versus Female Frontal Bone and Orbital Skeletal Morphology. pmc.ncbi.nlm.nih.gov/articles/PMC12459598
  7. Facial bone aging: an update and literature review (orbital enlargement, maxillary height loss, mandibular angle increase). www.sciencedirect.com/science/article/pii/S2352587826000185
  8. Rohrich & Pessa (2007). The Fat Compartments of the Face. Plast Reconstr Surg. pubmed.ncbi.nlm.nih.gov/17519724
  9. Microcephaly measurement in adults (n=2,508; head circumference M 56.3 ± 2.6 / F 54.2 ± 2.4 cm; correlation with brain volume and height). pmc.ncbi.nlm.nih.gov/articles/PMC9126574/
  10. Cephalic index - dolichocephalic / mesaticephalic / brachycephalic classification. en.wikipedia.org/wiki/Cephalic_index
  11. An Anthropometric Study of the Facial Index (facial index classes; M 88.15 / F 85.74). pmc.ncbi.nlm.nih.gov/articles/PMC13399996
  12. Cheekbone prominence and midface width ratios (bitemporal/bizygomatic 75-85%; bigonial/bizygomatic 70-82%). facialharmonyai.com/blog/cheekbone-prominence-bizygomatic-ratio
  13. Historical Tools of Anthropometric Facial Assessment: the neoclassical canons (systematic review). Aesthetic Surgery Journal 2022. academic.oup.com/asj/article/42/1/NP1/6369423
  14. Geniole, Denson, Dixson, Carré & McCormick (2015). Evidence from Meta-Analyses of the Facial Width-to-Height Ratio as an Evolved Cue of Threat. PLOS ONE (sex difference d = .11, n = 10,853; fWHR related to BMI, r = .31). pmc.ncbi.nlm.nih.gov/articles/PMC4504483/
  15. Kramer (2015). Facial Width-to-Height Ratio in a Large Sample of Commonwealth Games Athletes. Evolutionary Psychology (n = 3,481; men 1.98, women 1.92, d = 0.31; difference not significant after controlling for body size). pmc.ncbi.nlm.nih.gov/articles/PMC10430000/
  16. Tracking sexual dimorphism of facial width-to-height ratio across the lifespan: implications for perceived aggressiveness. Royal Society Open Science (2022; 17,607 passport images; male larger only in young adulthood, reversed after age 48). doi.org/10.1098/rsos.211500