What is the looksmaxxing test and what does it measure?
The RealSmile looksmaxxing test measures 17 facial-geometry metrics — symmetry, canthal tilt, FWHR, midface ratio, facial thirds and more, typically 15 of which a photo supports — from a single selfie as a Measurement Map of your features (real readings, not an attractiveness rating; Grammer & Thornhill 1994; Farkas 1994; Carré & McCormick 2008), and returns your validated Face Score free — a percentile cross-validated at r ≈ 0.8 against 1,566 human-rated faces. Your photo is processed and deleted right away. Nothing is kept unless you choose to save it.
- 68-point landmark detection · Private by default
- 17 distinct measurements — a map of your features, not a single black-box score
- Free baseline; 5-page softmaxxing report at /face-report
- Validated Face Score free with every scan (r ≈ 0.8 vs human ratings)
What is looksmaxxing?
Looksmaxxing is the systematic, evidence-based improvement of physical appearance by optimizing the three structural predictors of facial attractiveness supported by meta-analysis, symmetry, averageness, and sexual dimorphism (Rhodes, 2006), plus controllable presentation factors such as grooming, skincare, fitness, posture, and photo technique.
The term originated in online self-improvement communities and has gone mainstream on TikTok, Reddit, and YouTube as people realize that small, consistent changes to grooming, skincare, fitness, and photo presentation have outsized effects on first impressions.
First impressions of attractiveness, trustworthiness, and competence form from a face in roughly 100 milliseconds (Willis & Todorov, 2006), and a lot of it is under your control.
The looksmaxxing approach is divided into two categories. Softmaxxing includes non-invasive improvements like skincare routines, hairstyle optimization, eyebrow grooming, fitness, posture correction, and learning to photograph well. Hardmaxxing involves surgical or medical procedures. This looksmaxing test focuses entirely on softmaxxing, actionable improvements anyone can make: one photo returns your Face Score and your strongest measurement, free and without an account.
The goal of this test isn't to rate your attractiveness in any absolute sense. Instead, it gives you objective facial measurements compared to population averages, then tells you what you can actually improve through grooming, fitness, and photo technique.
Most looksmaxxing tests are golden-ratio "beauty calculators" that never check their looksmaxxing rating against how real people actually rate faces. RealSmile is the only one that does.
Your Face Score — free with every scan — is a validated percentile: it agrees with averaged human ratings at r ≈ 0.8, cross-validated against 1,566 human-rated faces in the SCUT-FBP5500 academic benchmark (60 ratings per face). See the full validation →
Is this looksmax test accurate — can an app really rate your looks?
This is the question that actually matters when you take a looksmax test, and it deserves a real answer rather than a marketing one.
Most looksmaxxing calculators are golden-ratio tools: they compare your proportions to a fixed template and never check whether that template matches how real people rate real faces.
On the standard academic benchmark, that golden-ratio approach reaches only about r ≈ 0.55–0.65 correlation with human ratings — which is why a "beauty calculator" can hand you a harsh looksmaxxing rating no actual person would agree with.
RealSmile's Face Score is measured differently. We ran our production pipeline against SCUT-FBP5500 — 5,500 face photos, each independently rated for attractiveness by 60 people — using 5-fold cross-validation, so every number is scored on faces the model never saw while fitting.
The Face Score reaches a cross-validated r ≈ 0.8 against the averaged human ratings (n = 1,566 held-out faces). For reference, the best published deep-learning models on this benchmark reach r ≈ 0.85–0.90.
We hold the geometry to the same test. Our original 17-metric blend failed it — no positive correlation with human ratings (r ≈ −0.33, slightly anti-correlated) — so we published that finding and stopped presenting the 17 metrics as an attractiveness number.
That is why the geometry is a Measurement Map, not a looksmax verdict. Our published evaluation identifies the test population and its limits so you can judge what the score supports.
| Method | r vs human ratings |
|---|---|
| Golden-ratio "beauty calculators"A fixed template, never checked against how real people rate real faces — can hand you a harsh rating no actual person would agree with | ≈ 0.55–0.65 |
| Our original 17-metric blendFailed the test (slightly anti-correlated), so the 17 metrics are presented as a Measurement Map, not a looksmax verdict | ≈ −0.33 |
| RealSmile Face Score5-fold cross-validated on SCUT-FBP5500 — 1,566 held-out faces, 60 raters each — scored on faces the model never saw while fitting | ≈ 0.8 |
| Best published deep-learning modelsThe ceiling on the same benchmark, for reference | ≈ 0.85–0.90 |
What your looksmax score actually means
A looksmax score is a position, not a verdict. Your scan gives you two different outputs, and reading them correctly is the whole point:
Your Face Score — the number to trust
A validated percentile that tells you where a still photo of your face lands against a large population. This is the output built to answer "how do I rate", because it is cross-validated at r ≈ 0.8 against how real people rate faces.
A percentile is a position, not a judgment of your worth — and most people who take a looksmaxxing test land closer to the middle than they fear.
The 17-metric Measurement Map — features, not a rank
Canthal tilt, FWHR, symmetry, facial thirds, midface ratio and more (17 calculated; a photo typically supports 15): a map of how each feature measures against population norms. It is not an attractiveness rating. We tested blending the 17 metrics into a single number and it did not predict how people rate a face (r ≈ −0.33), so we present it as a map.
Use it to find the one or two metrics scoring below their ideal range — the specific, usually softmaxxable things worth your attention. What each of the 17 measurements means →
A looksmax score is a position, not a verdict.
The honest read for almost everyone who takes this looksmax test: your looks are probably more average — in the statistical, middle-of-the-distribution sense — than your inner critic claims, plus one or two specific things you have been over-weighting. That is a far more useful, and more hopeful, answer than a single looksmaxxing rating out of 100.
Looksmax score — AI vs real people, on the same photo
When people take a looksmax test, half of them want a validated number and half want real people to weigh in. RealSmile is the only tool that gives you both on the same photo.
Your Face Score is the validated AI read (r ≈ 0.8 vs human ratings); your Human Score is the average of 10+ real people rating your photo — free, numbers only, moderated, and deletable anytime.
Put them head-to-head and the comparison is genuinely useful. If the AI and the crowd both land you in the same band, that is a strong, stable signal about how a still photo of you reads.
If they diverge, it usually points at something specific — an expression or lighting quirk the model reads differently than a person does — worth fixing before you use the photo for dating apps or LinkedIn. No golden-ratio looksmaxxing calculator can offer this, because none of them have a validated number or a real crowd behind it.