Golden Ratio Face Calculator: How Close Are Your Proportions to φ?
See each of your facial ratios next to the golden ratio and how far it deviates — as a table you can check, not a mystery score.
- Photo never leaves your device
- Every formula published
- Same photo, same result
- 1Add a photo
- 2See your ratios free
- 3Full report $9.90 · optional
Use a straight-on, evenly lit photo
Hair off the forehead, neutral expression, no glasses.
Analysed on your device. Your photo never leaves your device.
Three ratios compared with φ = 1.618
Mouth against nose width, the lower face, and the two lips — measured on MediaPipe’s reference face, not a real person. Your photo gets the same spans, values and deviations.
Mean deviation from φ
10.1%
Over the three scored ratios. Lower is closer to 1.618.
-
Mouth width ÷ nose width1.543
4.7% from φ
-
Nose base to chin ÷ lips to chin1.424
12% from φ
-
Lower lip ÷ upper lip height1.399
13.5% from φ
-
Face length ÷ face width1.152
not scored needs the hairline
What your golden ratio result shows
The golden ratio, written φ (phi), is the number 1.618…, the value of (1 + √5) ÷ 2. Two lengths are “in the golden ratio” when the longer divided by the shorter equals φ. This calculator takes three pairs of lengths on your face, divides each pair, and shows how far each result sits from 1.618. Nothing is rounded into a verdict: you see the ratios themselves.
Golden ratio part score
A number from 0 to 100 for this one part. It starts at 100 and loses 1.5 points for every 1% of mean deviation from φ. It is one of four parts of the overall score.
Mean deviation from φ
The average, over the three scored ratios, of how many percent each one is away from 1.618. Above or below φ counts the same, so the deviations cannot cancel each other out.
Ratio table
For each ratio: your value, φ, and the percent difference. The fourth row, face length ÷ width, is shown for interest and marked not scored.
Beside the table you get a confidence value with one decimal (for example 72.6%) and plain notes on head angle, light and the hairline. The annotated copy of your photo is drawn on your device only. If the head is turned more than 10° or tilted too far up or down, the card asks for a retake and withholds the scores: a turned head shortens the mouth and nose widths unevenly, and that changes these ratios more than real differences between faces do.
Where the “golden face” idea came from
Artists and anatomists have compared faces with proportional rules for centuries; the “neoclassical canons” of equal thirds and fifths are the best known. The specific claim that facial harmony follows φ is more recent. It is tied to the Marquardt “Phi mask”, a template of decagon-based lines laid over a face, which its author described in an orthodontic journal interview (JCO, 2002). That interview is the primary source of the claim and is not peer reviewed, so we cite it as a claim, not as evidence.
Two peer-reviewed studies tested the idea. Holland (2008) compared the mask with faces and concluded that relying on fashion-model faces and φ does not describe a wide range of faces. Pallett, Link and Lee (2010) asked people to adjust face layouts and found that the preferred proportions were not the φ-based ones: the preferred distance between the eyes and mouth was about 36% of face length and the distance between the eyes about 46% of face width, which they called “new golden ratios”. Both results point the same way: φ is an interesting reference value, not a measured norm.
Population studies add a second caution. Farkas and colleagues measured thousands of faces and found that average proportions differ between ethnic groups and between individuals (Farkas 1985, 2005). A single number cannot describe all of that. So this page reports how close your photo’s geometry is to 1.618, and stops there. If you want a golden ratio face calculator that tells you a face is or is not “golden”, this is not it; the tool measures, and the interpretation stays with you.
The ratios, landmark by landmark
Every length is measured between MediaPipe landmarks after the eye line is levelled and a small head turn is undone, and ratios are unitless, so photo size and pupil distance cancel out. Three ratios are scored. A fourth is displayed but excluded from the score.
| Ratio | Landmarks | How it is calculated | Canonical face |
|---|---|---|---|
| Mouth width ÷ nose width | 61↔291 ÷ 64↔294 | Distance between the mouth corners divided by the distance between the outer edges of the nostrils (nose wings). | 1.54 |
| Nose base to chin ÷ lips to chin | 2→152 ÷ lip line (13/14)→152 | Vertical span from the base of the nose to the bottom of the chin, divided by the span from the lip closure line to the bottom of the chin. | 1.42 |
| Lower-lip height ÷ upper-lip height | 14→17 ÷ 0→13 | Lower lip from the inner lip line to its bottom edge, divided by the upper lip from its top edge to the inner lip line. | 1.40 |
| Face length ÷ face width (not scored) | 10→152 ÷ 234↔454 | Upper forehead to chin divided by the width between the cheek edges. Shown, never used in the score. | 1.15 |
“Canonical face” means MediaPipe’s own average face model, the same reference we use to test every measurement. It is a calibration point, not a target and not a typical person: your values may sit above or below these numbers for ordinary reasons, including expression, since a smile widens the mouth and shortens the lips.
Why face length ÷ width is shown but not scored
Classic golden-face templates measure face length from the hairline. The model cannot see the hairline: hair, lighting and forehead shape make it invisible to the landmarks. We therefore measure length from the upper forehead (landmark 10), which is an estimate, and the canonical value from that point is 1.15, far from 1.618. Scoring a ratio whose top end we cannot locate would mix photo noise into the number, so it stays on the card for reference only. Face length ÷ width is also used by the face shape detector, where it has a published decision table.
Why only three ratios are scored
The scored ratios are the ones whose endpoints sit on clearly visible edges in a front photo: mouth corners, nostril edges, nose base, lip line and chin. Ratios that depend on the hairline or on profile features cannot be measured reliably from one front photo. Features such as the nasofrontal and nasolabial angles are listed as Not scored — needs a side photo in the full report rather than being guessed.
How deviation and the part score are calculated
Four short steps, all of which you can repeat with a calculator.
| Step | Formula | Example (canonical face) |
|---|---|---|
| 1. Ratio | length A ÷ length B | 1.543 / 1.424 / 1.399 |
| 2. Deviation of each ratio | |ratio − 1.618| ÷ 1.618 × 100 | 4.7% / 12.0% / 13.5% |
| 3. Mean deviation | average of the three deviations | 10.1% |
| 4. Part score | 100 − 1.5 × mean deviation, limited to 0–100 | 84.9 |
Why golden ratio weighs only 15% of the overall score
The overall score is symmetry 35%, facial thirds 25%, facial fifths 25% and golden ratio 15%. The golden-ratio part has the smallest weight because its evidence base is the weakest of the four: the Phi mask has been questioned by Holland (2008), and the preferred proportions in Pallett et al. (2010) differ from φ. Even MediaPipe’s average face model, a face built to be a neutral reference, sits about 10% from φ on these three ratios, so a part score near 85 is what an average geometry produces. You can see the full weighting on the face rating page.
The “% off” column is relative to φ, not to a norm. A ratio 12% below 1.618 is not an error. It means the number is 12% away from a mathematical constant that was never established as a population average.
How to read your ratio table
Read the rows before the part score. Each row tells you which pair of lengths is nearer to or further from 1.618, and by how much; the part score only averages them.
| What you see | What it usually means |
|---|---|
| One row far from φ, the other two close | The overall mean is driven by one pair of lengths. Check that photo’s expression first: a smile moves the mouth corners and lips. |
| All three rows on the same side of φ | Common. The canonical face is below φ on all three ratios, so the direction alone says little. |
| Values change a lot between two photos | The photo, not the face, is changing: lens distance, head angle or expression. Compare photos taken the same way. |
| Low confidence next to the table | Light, size or pose limited the landmark placement. Treat the deviations as rough. |
| Lip ratio differs most | Lip heights are small distances, so a one-pixel landmark shift changes this ratio more than the others. |
Overall, the table describes the geometry of one photograph relative to a mathematical constant. It says nothing about how a face is perceived: research summarised by Rhodes (2006) and Little et al. (2011) finds that averageness, symmetry and other cues matter, and that no single ratio explains the result. If you want the other measures, the face symmetry test mirrors the landmarks across your midline, and the jawline analysis covers the lower face from a front view.
Getting ratios you can trust
Relax the mouth
Two of the three scored ratios use the mouth or lips. A neutral, closed mouth gives repeatable lip and mouth-width values; a smile does not.
Keep the camera at eye level
A camera below the chin lengthens the lower face and changes the nose-to-chin ratio; above it does the opposite. Eye level keeps vertical spans honest.
Stand back, then crop
A phone held close enlarges the nose relative to the mouth because it is nearer the lens. An arm’s length or more reduces that distortion.
Look straight at the lens
Head turns beyond 10° are refused, and smaller turns are only partly corrected. Facing the lens keeps nose and mouth widths comparable.
Use even light
Hard shadows around the nostrils and lips move the landmarks that define these ratios. Soft light from the front lets the model place them cleanly.
Repeat, then compare
If you want to know whether a number is stable, take a second photo the same way. Small differences between photos are normal; treat only repeated patterns as informative.
More free tools, same photo
Frequently asked questions
What is the average golden face ratio?
The reference value is φ = 1.618. This tool divides three pairs of face measurements and compares each result with it: mouth width ÷ nose width, nose-base-to-chin ÷ lips-to-chin, and lower-lip ÷ upper-lip height. Pallett et al. (2010) found that preferred proportions in their experiment averaged close to 0.36 and 0.46 rather than φ-based values.
How to calculate golden ratio of face?
Divide one measurement by another, then compare the result with 1.618. Deviation % = |ratio − 1.618| ÷ 1.618 × 100. Here the three scored ratios are averaged, and the part score is 100 − 1.5 × that mean deviation, limited to 0–100. The table on this page lists the landmarks behind every ratio so you can repeat the arithmetic.
Is a golden ratio face considered beautiful?
The evidence is mixed. Marquardt’s mask claimed that φ describes beauty, but that claim comes from an interview, not a peer-reviewed study. Holland (2008) found the mask does not describe a wide range of faces, and Pallett et al. (2010) found preferred proportions that differ from φ. Treat φ as a reference value, not a standard.
What is a good golden face ratio score?
No scientific threshold exists. On this page the part score is 100 minus 1.5 points for every 1% of mean deviation from φ, so a mean deviation of 10% gives 85. MediaPipe’s average face model lands near that value, which means ordinary faces rarely reach 100. The number describes closeness to φ, nothing more.
Is the Golden Ratio Calculator suitable for all ethnicities and skin tones?
The arithmetic is the same for everyone, but the reference is not neutral. Farkas et al. (2005) measured clear differences in facial proportions between ethnic groups, so one fixed target cannot describe all populations. Landmark detection can also be less steady in poor or uneven light. This is why the golden-ratio part carries only 15% of the overall score.
Is the golden ratio face test biased toward certain looks?
Any single target value favours the proportions it encodes. φ-based masks were drawn from a narrow set of faces, as Holland (2008) discusses, and population norms differ (Farkas 2005). That is why we show the raw ratios and each deviation instead of a verdict, and keep the part at the lowest weight in the score.
Method and references
Landmarks come from Google’s MediaPipe Face Landmarker (478 points including the irises), run in your browser in single-image mode on the CPU, so the same photo always gives the same numbers. The three scored ratios and the factor of 1.5 points per 1% mean deviation are our own engineering choices, published above; the weight of 15% follows a common public scoring scheme. Canonical values (1.54, 1.42, 1.40 and 1.15) are measured on MediaPipe’s canonical face model. Face length is measured from the upper forehead because the hairline is not visible to the model.
- Holland E. (2008). Marquardt’s Phi mask: pitfalls of relying on fashion models and the golden ratio to describe a beautiful face. Aesthetic Plastic Surgery 32:200–208. doi:10.1007/s00266-007-9080-z
- Pallett PM, Link S, Lee K. (2010). New “golden” ratios for facial beauty. Vision Research 50(2):149–154. doi:10.1016/j.visres.2009.11.003
- Farkas LG, Hreczko TA, Kolar JC, Munro IR. (1985). Vertical and horizontal proportions of the face in young adult North American Caucasians: revision of neoclassical canons. Plastic and Reconstructive Surgery 75(3):328–337. doi:10.1097/00006534-198503000-00005
- Farkas LG, Katic MJ, Forrest CR. (2005). International anthropometric study of facial morphology in various ethnic groups/races. Journal of Craniofacial Surgery 16(4):615–646. doi:10.1097/01.scs.0000171847.58031.9e
- Marquardt SR. (2002). JCO interviews Dr. Stephen R. Marquardt on the golden decagon and human facial beauty. Journal of Clinical Orthodontics. Primary source of the mask claim; an interview, not peer reviewed. jco-online.com
- Rhodes G. (2006). The evolutionary psychology of facial beauty. Annual Review of Psychology 57:199–226. doi:10.1146/annurev.psych.57.102904.190208
- Little AC, Jones BC, DeBruine LM. (2011). Facial attractiveness: evolutionary based research. Phil. Trans. R. Soc. B 366:1638–1659. doi:10.1098/rstb.2010.0404
Your photo never leaves your device
The face model is downloaded to your browser and runs there. Your image is never sent to our server or any third party — this site has no server-side code or upload endpoint at all, and its security policy lets the page contact only this site and Cloudflare's cookieless page-view counter. Read the privacy policy.