Formula shown · pps-v2.0.0
How the Score Is Calculated (Formula Shown)
Two layers, two different confidence levels
Most tools in this category publish a single number and hide how it was made. We split the read in two, because the two halves deserve different levels of trust.
Layer 1 — Photo Presentation Score (measured)
PPS = 0.22·lighting + 0.18·sharpness + 0.16·framing + 0.14·background + 0.12·contrast + 0.10·composition + 0.08·colour
Each term is computed from pixel statistics inside and around the detected face region: a 3×3 luminance grid for light evenness, normalised Laplacian variance for sharpness, face-height-to-frame-height for framing, edge energy outside the face box for background quiet, the p5–p95 tonal spread for contrast, headroom position for composition, and the distance of the background from neutral for colour cast — the background, because skin is genuinely chromatic and judging cast against the face made the metric unscoreable for everyone. These are properties of the photograph, which is why they are the part you can fix in ten minutes.
| Measured | Weight | Computed from | What it is not |
|---|---|---|---|
| Light evenness | 22% | Spread and left/right difference across a 3×3 luminance grid | A judgement about your skin |
| Sharpness | 18% | Laplacian variance normalised by local contrast | A measure of camera quality alone |
| Framing & crop | 16% | Face height as a share of frame height, plus off-centre penalty | A rule about how you should look |
| Background quiet | 14% | Edge energy outside the face box | A judgement about your home |
| Contrast & tone | 12% | p5–p95 luminance spread inside the face box | A colour-grading opinion |
| Headroom & centring | 10% | Distance of the head top from the ideal 16% line | A composition dogma |
| Colour cast | 8% | Distance of background Cb/Cr from neutral (grey-world) | A skin-tone judgement |
Skin tone — measured with ITA°, never labelled as identity
Skin lightness is measured the way dermatology measures it: ITA° = atan((L* − 50) / b*) × 180/π, computed from the mean CIE L*a*b* of your skin pixels after white-balancing against the light estimated from the background. The result is shown as one of six published bands (very light, light, intermediate, tan, brown, dark) and a three-way warm/neutral/cool hue axis. A phone photo is not a spectrophotometer, so the band is coarse on purpose — and when the colour cast is too strong to correct, the engine refuses to report a band at all rather than guess.
This is a measurement of pixels, not a classification of people. There is no ethnicity field in the engine, in the API or in the database; the scan endpoint accepts six fixed tone codes, three fixed hue codes and bounded numbers, and nothing else. If you want your read tailored to how you identify, you tell us — a single selected option, never an inference from your face.
Layer 2 — Score (estimated)
Score = 0.22·D_sym + 0.20·D_gold + 0.18·D_thirds + 0.14·D_fifths + 0.14·D_eyes + 0.12·D_jaw
Each D is an integer 0–100 estimate. The result is rounded to a whole number out of 100 and also shown to one decimal place out of 10. We do not display fabricated decimals beyond that.
| Signal | Weight | Derived from | What it is not |
|---|---|---|---|
| D_sym · Symmetry | 22% | Mirror difference between left/right halves of the detected face region | A clinical symmetry measurement |
| D_gold · Golden Proportion | 20% | Face height/width placement inside a broad reference band | An aesthetic ideal or a phi "score" |
| D_thirds · Facial Thirds | 18% | Balance of three vertical luminance/edge bands | Landmark-based anthropometry |
| D_fifths · Facial Fifths | 14% | Consistency of five horizontal edge-density bands | A five-eye rule measurement |
| D_eyes · Eye Area | 14% | Upper-band edge density and spacing proxy | Canthal tilt from landmarks |
| D_jaw · Jawline | 12% | Lower-band width taper proxy | Gonial angle (not scored) |
Bands, not grades
Each estimate is reported as a band: 80–100, 60–79, 40–59, or below 40. Scores are placement inside a broad reference band, not grades. 100 is not the goal. A broad band is used precisely because the input is coarse. The Score is a weak geometric signal. It does not predict matches, dates, replies, or your worth.
The confidence band
Every score ships with a computed ± range, not an asserted one. Penalties are added for a short side under 800 px (+1) or under 480 px (+2), soft focus (+2), uneven light (+1), a face smaller than 30% of frame height (+1) and a possible head turn (+1). The total decides the label:
| Penalty | Confidence | ± on the score | What it means |
|---|---|---|---|
| 0–1 | high | ±5 | Front-on, sharp, evenly lit, face fills the frame |
| 2–3 | moderate | ±8 | One or two conditions are working against the read |
| 4+ | low | ±14 | Treat this as a rough direction, not a number |
This is why the report shows a range such as 58–74 rather than a single figure that looks more precise than it is.
Market fit
Chapter 30 of the paid report ranks four dating markets for the same profile, using a published rule:
fit = 0.34·goal alignment + 0.28·photo readiness + 0.20·effort fit + 0.18·tool fit
Goal alignment is 100 when the market matches your stated goal and 55 when it is adjacent. Photo readiness is your measured PPS score, or a self-report band if you skipped the pixel scan. Effort fit compares the time you have each week with what the market asks for. Tool fit compares your main app with the apps that market runs on. Fit is a positioning score. It does not forecast matches, and it does not change who you are.
What happens in your browser
- Your photo is drawn into a canvas, resized to at most 1600 px and re-encoded — this strips EXIF metadata including GPS.
- A skin-tone projection locates a plausible front-facing face region. If none is found, or two separated skin clusters suggest more than one face, the tool says so instead of inventing a score.
- Pixel statistics are computed and mapped onto the seven measured properties and the six estimated bands. There is no randomised jitter: the same pixels produce the same numbers.
- Only numbers (never pixels) are posted, so a result link exists for up to two hours. Free scans leave no image behind.
Hard limits
- Estimated with a lightweight on-device geometry heuristic — not a 478-landmark clinical measurement. We do not use MediaPipe's 478 landmarks, any trained face model, or any clinical measurement.
- A single front-facing photo cannot support side-profile claims. Where that is true, we say "not scored" instead of guessing.
- This tool measures the photo, not your face directly. Lighting, angle, focal length and expression all move the reading.
- Aesthetic reference ranges in the literature carry documented bias. We treat them as rough bands, not ideals.
- We never rank you against an ethnicity-based ideal or against another person's face.
- The Score is a weak geometric signal. It does not predict matches, dates, replies, or your worth.
Reproducibility and versions
The same photo with the same pps-v2.0.0 produces the same result. If we change the heuristic, we change the version string and say so — we never silently re-score your face. Disclaimers version: dis-v1.0.
Hard limits, in plain language
There is no leaderboard here. We don't rank you against other people. Your free result is never blurred, locked, or held back. The Score is a weak geometric signal. It does not predict matches, dates, replies, or your worth.
Sources behind the framing
Background reading: Langlois & Roggman (1990); aggregated swipe-data analyses. Directional context, not a validation of this tool. These are background references on why facial preferences are noisy and biased; they are not validation of our score.
- Langlois & Roggman (1990), Psychological Science — averaged faces.
- Rhodes (2006), Annual Review of Psychology — symmetry and averageness, and their limits.
- Willis & Todorov (2006), Psychological Science — rapid impression formation.
- Little, Jones & DeBruine (2011) — individual variation in preference.
- Farkas (1994), Anthropometry of the Head and Face — population-specific norms.
- Lugaresi et al. (2019), MediaPipe — cited only to name what we do not do.
Non-negotiable
- Entertainment & self-insight only. Not medical, psychological, or dermatological advice.
- The Score is a weak geometric signal. It does not predict matches, dates, replies, or your worth.
- If a low score is making you feel bad, close the tab. This is a measurement, not feedback that should affect how you see yourself.
- No procedures, surgery, fillers or injectables are recommended anywhere on this site.
- Face Rate is for adults 18 and over.