Understand your face. Then decide what, if anything, to change.

What people come here to work out

• Work out whether the thing I keep noticing is actually unusualIt is measured against a population you name, with the error bar printed next to it.
• Find out what would change anything, and what would change nothingEvery recommendation carries an evidence grade, a cost band and a time to visible effect.
• Stop being sold a procedure by the person who performs itWe take no commission, no referral fee and no affiliate revenue from anything we mention.
• Know whether something I already did made a differenceThe next analysis is compared to the last one on the same axes, with the same error budget.

What the evidence actually says

Every card below is one finding, hedged the way its own source hedges it, with the over-reading it invites named and refused.

15 of 15 cards are withheld. A card may not be published until an independent reader has checked its source against the primary paper. Until then we would be rating a claim on the strength of a paper nobody here has opened, which is the same defect as asserting one. How we cite.

4 of these cards exist in order to refuse a popular claim rather than to make one. Refusing publicly is content, so they are published as refusals — under the same verification rule as the rest.

Attractiveness produces small-to-moderate advantages in situations where other people form quick impressions, and little or no advantage where objective performance can be measured. Most of this research is correlational; no study has shown that changing someone's appearance causally changes their life outcomes. A substantial share of the measured "beauty premium" reflects grooming, confidence and communication — all modifiable — rather than facial structure.

Three ways to answer the same question

Fixating on one feature, asking an app for a number out of ten, or measuring.

What your analysis is actually personalised on

Every one of these changes a number in your report, not a word in it.

• Reference population

  You declare your ancestry, sex and age band, and your percentiles are computed against norm tables for that group. We never infer it from your photographs.

• What your photographs will support

  A measurement whose confidence interval is too wide on your images is suppressed rather than reported as a percentile.

• Health, medication and contraindications

  Anything unsafe for you is removed from the plan before you see it, not flagged after.

• Budget and time

  The plan is banded by cost and by time to visible effect, so it is a plan you could actually follow rather than a wish list.

What is in the report

• 181 biometric tests across 13 facial regions, each with its confidence interval
• 7 trait axes reported separately, each with its evidence strength stated
• Percentiles against a reference population we name, or the raw measurement where we have no norms
• An evidence-graded plan with cost bands, time to visible effect and a risk band per item
• The name and credential of the person who reviewed your analysis
• Every measurement we suppressed, and why

How it goes

1. Answer the intake

   Goals, routine, health, and the wellbeing screen.

2. Take the guided photographs

   On your phone, with an overlay and a distance meter.

3. We measure, and a person reviews it

   The measurements are computed, then checked and signed by a named reviewer. 48 hours.

4. You get the report and the plan

   Yours to keep, print and take to a clinician.

Seeing the size of a change before you make it

Your own pixels, moved. A visualization is a deterministic warp of your photograph by the size of a change we described — not a generated face, and not a prediction of a result.

• Every render is paired with the unmodified photograph it came from.
• An identity check runs on the output: if it stops looking like you, we do not show it.
• Rendered images are watermarked in the pixels and carry their provenance in the file.
• We will not render an outcome we could not describe as a measurement.

How the renders are produced, and what we refuse to render

Measuring whether anything actually changed

Your next analysis is compared to your last one, on the same axes, with the same error budget.

• Symmetry
• Averageness
• Dimorphism
• Neoteny / visual age
• Adiposity
• Homogeneity
• Proportionality

Every tracked number is drawn with its confidence band. A change smaller than the band is not a change, and we will say so rather than let you read a trend into noise.

Every measurement, by region

The count is the sum of this table. It is checked against the specification on every build.

Why two reports do not read the same

• Your percentiles depend on a group you chose

  The same measurement is a different percentile against a different reference population, and we tell you which one you are being compared to.

• Your suppressed measurements are yours

  Which tests we could not report honestly depends on your photographs, so no two reports contain the same set.

• Your plan is filtered by your own contraindications

  Two people with the same measurements get different plans if one of them is pregnant, or on isotretinoin, or allergic to something.

• A person signed it

  A named reviewer can change or remove anything in your report, and their name is on the front of it.

What the system does, and what it will not do

• Measurement

  521 landmark points with a solved perspective camera, so measurements are taken on a reconstructed mesh rather than on pixels.

  How this works

• Reference populations

  Percentiles computed against sex, ancestry and age band, with the source of every norm table published.

  How this works

• Human review

  Every analysis is checked by a trained reviewer whose name and credential appear on your report.

  How this works

• What we refuse to build

  We do not produce an attractiveness score. Every publicly available training set for that task is both non-commercially licensed and demonstrably biased by ethnicity.

  How this works

Your measurements are compared to a population you choose

A percentile is meaningless without a reference group. "The 70th percentile" is a different statement depending on who the other 99 people are.

So we ask you to declare your ancestry, and we compute your percentiles against reference data measured on that population — not against a single global standard. The neoclassical proportion canons that most facial-aesthetics content is built on hold in 0% to 33% of real people even in the populations they were derived from, and they were derived from European faces.

Where we don't have good normative data for your group, we say so and show you the raw measurement instead of inventing a percentile. That gap is real, it is worst for South Asian, Middle Eastern / North African and Latin American reference data, and we publish which tables we have and where they came from.

We do not infer your ancestry from your photographs. You tell us, or you skip the question and get measurements without percentiles.

See every normative table we use

Bozkir et al. 2004; Khoshab et al. 2022; Farkas, Katic & Forrest 2005

The mechanics

Six guided photographs, taken on your phone, with the camera geometry solved rather than assumed.

• Frontal, neutral
• Frontal, smiling
• Left profile
• Right profile
• Left three-quarter
• Right three-quarter

1. An overlay and a distance meter guide each view, so the camera geometry is the same every time.
2. Each photograph is checked for focus, exposure, expression and pose before it is accepted.
3. The focal length is solved rather than assumed, because a phone selfie reads a nose as longer than it is.
4. Landmarks are fitted in three dimensions, and every measurement carries the error budget of that fit.
5. A trained reviewer checks the result, and can change or remove anything before you see it.

What happens after the report

• Care

  A real person, in a thread that has your analysis in front of it. Not a widget from another company.

• The analyst's letter

  The reviewer writes to you about what they changed and what they were unsure about.

• Tracking that outlives the subscription

  Reports you have already had stay readable whether or not you renew.

Questions people ask before they buy

Disclaimers

1. Beauty Privilege does not provide licensed clinical or medical services. Nothing here is a diagnosis or a treatment recommendation for a medical condition.
2. Most of the changes we recommend are achievable without surgery. Results vary between individuals, and several recommendations take months to show any visible effect.
3. Images marked "Digitally rendered" are computed modifications of a real photograph, produced to illustrate the size of a possible change. They are not photographs and not predictions of a surgical outcome.
4. We do not produce an attractiveness score or rating. Percentiles are stated relative to a named reference population and are not a judgement of how attractive anyone is.
5. We do not accept customers under 18, and we stop and refund anyone whose wellbeing screen indicates that appearance concerns are causing significant distress or taking up significant time.