What is canthal tilt?

Canthal tilt is the angle between the line from the inner eye corner to the outer eye corner and the pupil line. Positive means the outer corner sits higher on screen, negative means it sits lower, and the test prints the angle for each eye, their mean and the gap between them. The informal term hunter eyes is not a measurement category here: this page reports one angle, not an eye-shape label.

What does a positive canthal tilt mean?

It means the outer corner of the eye sits higher than the inner corner, on average across the two eyes. In the ordinary samples on this page the majority sit above 0°: 93.7% of the Western sample and 98.3% of the East Asian sample. It is a common shape, not a rare feature, and the label is not a judgement about how a face looks.

Is positive canthal tilt better than negative?

There is no consensus. The studies in the dated table report associations and sample means; none supplies a cut-off, and this page does not score. The head page's canthalTilt row does score against a performer band, but that is index calibration, not an aesthetic conclusion.

Why are my left and right angles different?

The median left-right difference is 0.85° in the Western sample and 0.75° in the East Asian sample, and at p90 the gap is 2.09° and 2.21°. A difference of 2–3° is still inside the ordinary range of these samples. Faces are not mirror images. A sudden change, a drooping lid, double vision, pain or swelling needs a clinician.

Can makeup, eyeliner or contact lenses change the reading?

Yes. Eyeliner extends the corner, lashes can cover it, and colored or circle lenses change the iris position that the pupil line depends on. For a like-for-like comparison, take both photos with the same everyday eye condition, or both with a clean face.

Does head tilt or camera angle change the number?

Yes. The tool aligns the face on the pupil line and rejects photos outside its pose gates, but remaining gaze and pitch still move the reading. Look at the lens, keep the head straight and hold the camera at eye level.

Why is there no score on this page?

Three reasons: the literature has no consensus threshold; the error on a single 2D photo is larger than the differences a rank would separate; and a positive tilt is common, so a higher-is-better scale would not tell faces apart. The AI attractiveness test has an index if that is what you want.

How is this different from the AI attractiveness test?

The head page combines seven proportion rows into a 0–100 index; this page reports only the eye-corner angles. Here there is no total, no rank and no reference switch. Both pages use the same detector and the same 478 landmarks.

Is this a medical diagnosis?

No. This is a 2D landmark angle, not a clinical measurement of the palpebral fissure. Sudden asymmetry, a drooping eyelid, double vision, pain or swelling should be assessed by an eye or plastic-surgery clinician, not by a photo tool.

On-device · Free · No sign-up

Canthal Tilt Test: Is Your Tilt Positive, Neutral or Negative?

Measure the angle between the inner and outer corner of each eye. See left, right, mean and the difference — and why a positive tilt is the norm, not a rare trait.

This page measures the angle between each eye's inner and outer corner, relative to the pupil line, in degrees. It shows four readings — left eye, right eye, mean and the gap between them — and no combined score. In the two ordinary samples measured here, 93.7% of the Western sample and 98.3% of the East Asian sample sit above 0°, so a positive tilt is common rather than rare; common is not better. The photo is processed in your browser and never uploaded, and this is not a diagnosis.

How the angle is measuredFour readings, no combined score.

Your photo stays in your browser.

Runs on Google MediaPipe in your browser. Your photo never leaves your device.

How it works

01Add a photoUpload a file or take one with your camera. The photo stays in your browser.
02We find 478 pointsMediaPipe maps the points on your face, on your device.
03Read four anglesLeft, right, mean and the gap between them, printed as measured. No score.

What the Canthal Tilt Test Measures

The test reads the tilt of each eye separately. For each eye it draws a line from the inner corner to the outer corner, compares that line with the pupil line, and reports the angle in degrees.

Four numbers come out: left eye, right eye, the mean of the two, and the gap between them. Positive means the outer corner sits higher than the inner corner on screen; negative means it sits lower.

There is no combined score, no rank and no eye-shape category. A single photo describes one moment, under one camera and one head pose, and that is all the page claims.

It is not a diagnosis. The reading is a 2D landmark angle, not a clinical measurement of the palpebral fissure.

How the Eye-Corner Angle Is Calculated

MediaPipe Face Landmarker returns 478 points. Six of them define this reading.

The six MediaPipe landmarks used for the angle
Landmark	MediaPipe index	Used for
Left eye inner corner	133	Start of the left inner → outer line.
Left eye outer corner	33	End of the left inner → outer line.
Right eye inner corner	362	Start of the right inner → outer line.
Right eye outer corner	263	End of the right inner → outer line.
Left iris center	468	Left endpoint of the pupil line.
Right iris center	473	Right endpoint of the pupil line.

The whole face is first rotated until the pupil line — the line through the two iris centers — is horizontal. Each eye's corner line is then measured against that horizontal, so head roll does not enter the reading twice.

If either iris center cannot be read, there is no pupil line and no angle. The reading is marked unavailable rather than filled with a fallback, and the page asks for another photo. When only one eye can be read, the angle is also unavailable, because the mean and the gap both need two.

Positive means the outer corner is higher on screen. The sign is a convention, not a grade.

This is a 2D landmark angle. A clinical palpebral fissure inclination uses different reference lines, endpoints and equipment, so do not compare these numbers directly with published values.

Positive, Neutral and Negative

Positive: the mean angle is greater than +1°. Neutral: the mean sits within ±1° of level. Negative: the mean is below −1°.

The ±1° cut is a display threshold, not a clinical one. It was chosen because the median left-right difference in the two samples is 0.85° and 0.75°, so 1° is about the size of a repeat-measurement wobble on one photo. Other tools may use 0°, ±2° or another cut; compare numbers, not labels.

Neutral is a narrow class in practice: 6.76% of the Western sample and 1.50% of the East Asian sample fall inside ±1°. Most faces land on the positive side of the line.

A negative reading is a description, not a defect, and a positive reading is not a target. Neither label says anything about health or looks on its own.

Positive Is the Norm, Not a Rare Trait

Two convenience samples of ordinary stock portraits were measured with the same detector. Western: 1,301 scored photos, mean +4.97°, median +5.10°, middle 50% from +2.94° to +7.00°; 93.7% sit above 0° and only 3.7% below −1°. East Asian: 534 scored photos after removing photos shared with the Western pool, mean +6.55°, median +6.76°, middle 50% from +4.54° to +8.53°; 98.3% sit above 0° and 0.8% below −1°.

The head page's performer band for this row is 2.50°–6.50° in the Western profile and 4.90°–8.17° in the East Asian profile. Both sit almost exactly on top of the ordinary middle 50% above. In the Western sample 49.3% of faces fall inside the band; in the East Asian sample, 40.8%.

That overlap matters. The performer band is a calibration interval from a sample of performers, not a threshold that separates beautiful faces from ordinary ones. A positive tilt is the common case in both samples, so this page does not treat it as a rare or elite trait.

Ordinary samples beside the performer bands they overlap
Sample	n	Median	Middle 50%	Above 0°	Performer band	Inside band	Above band
Western ordinary portraits	1,301	+5.10°	+2.94° to +7.00°	93.70%	2.50°–6.50°	49.27%	30.21%
East Asian ordinary portraits	534	+6.76°	+4.54° to +8.53°	98.31%	4.90°–8.17°	40.82%	31.27%

What Published Research Finds

The claim under test is that a positive lateral canthal tilt is a distinctive sign of an attractive face. The nine studies below are ordered by first publication date; each row is marked as supporting, qualifying or challenging that claim.

Dated canthal tilt studies and what they bear on
Date	Study	Sample or method	Finding	Bearing on the claim
2007-01-01	Bashour and Geist, Is medial canthal tilt a powerful cue for facial attractiveness?Ophthalmic Plast Reconstr SurgDOI	Forced-choice photo pairs; medial canthal tilt digitally increased; women's faces	The version with increased medial canthal tilt was preferred in 93% of choices (p < 0.0001).	Qualifies the claim
2009-08-01	Price et al., Eyebrow and eyelid dimensions: an anthropometric analysis of African Americans and CaucasiansPlast Reconstr SurgDOI	89 African American and 75 white adults, 20–80 years; standard photographs	Palpebral fissure tilt differed by sex in both groups; eyelid and brow heights differed by group.	Qualifies the claim
2013-01-01	Vasanthakumar et al., Anthropometric analysis of palpebral fissure dimensions and its position in South Indian ethnic adultsOman Med JDOI	200 South Indian adults, 18–26 years; standard frontal photographs	Mean palpebral fissure inclination was 5.053° in men and 6.102° in women.	Challenges the claim
2015-01-01	Sforza et al., Comparison of soft-tissue orbital morphometry in attractive and normal Italian subjectsAngle OrthodDOI	199 normal and 126 attractive Italian adults, 18–30 years; 3D imaging	The attractive group had a larger palpebral fissure inclination, a shorter inner canthal distance and a longer fissure; orbital height increased only in women.	Supports the claim
2020-12-01	Lee et al., Diagnostic analysis of vertical orbital dystopia and canthal tilt for surgical correctionJ Korean Assoc Oral Maxillofac SurgDOI	76 Korean patients, mean age 23.12 years	Mean canthal tilt was 8.45° ± 3.53° on the right and 8.42° ± 3.81° on the left; the class I subgroup read 3.21° and 3.18°.	Qualifies the claim
2022-08-15	Liu et al., Age-related changes of the periocular morphology: a two- and three-dimensional anthropometry study in CaucasiansGraefes Arch Clin Exp OphthalmolDOI	68 younger and 73 older white adults; 2D and 3D imaging	Palpebral fissure inclination changed significantly with age (p ≤ 0.004); the inner canthus moved down and outward and the outer canthus down and inward.	Qualifies the claim
2023-06-05	Liu et al., Using artificial intelligence to quantify sexual dimorphism in aesthetic facesAesthet Surg J Open ForumDOI	21 women and 21 men, white celebrities aged 23–42; AI landmarking	Attractive women had a larger canthal tilt than attractive men.	Supports the claim
2024-12-30	Gao et al., Racial and sexual differences of eyebrow and eyelid morphology: three-dimensional analysis in young Caucasian and Chinese populationsQuant Imaging Med SurgDOI	46 Chinese and 101 white adults, 18–30 years; 3D imaging, 32 landmarks	Periocular shape differed by group and sex; white participants had a larger canthal tilt and outer canthal angle than Chinese participants.	Qualifies the claim
2026-09-12	Wang et al., Associations between palpebral fissure morphology and AI-derived facial attractiveness scores in Asian adultsJ Plast Reconstr Aesthet SurgDOI	3,517 Asian adults (SCUT-FBP5500); nine periocular measurements	Higher AI-attractiveness groups were associated with a larger canthal tilt, but brow–eyelid distance and inner canthal distance explained the most variance; human ratings and age-adjusted sensitivity analyses broadly agreed.	Qualifies the claim

What changed, what did not

Changed: measurement moved from 2D photographs and tape measures to 3D imaging; samples extended from Italian, South Indian, Korean, Chinese and white celebrity groups to a 3,517-person Asian set scored by an AI model; and the conclusion moved from a single beauty cue to a shape measure with age, sex and population variation.

What did not change: no study supplies a threshold for how positive a tilt should be; ordinary samples sit above 0° on average; the strongest supportive result tested the medial canthus, not the lateral line measured here; and none of the papers defines an eye-shape category.

Read the strongest supportive result carefully: it digitally increased the medial canthal tilt and tested women's faces in forced-choice pairs. It supports a preference for that manipulation, not for the outer-corner angle this page reports.

The 2026 Asian study is an association, not a cause. Attractiveness scores were AI-derived, the canthal tilt was not the strongest periocular term, and the authors' human-rating and age-adjusted sensitivity analyses broadly supported the pattern without turning it into a rule.

The 2015 Italian study compared an attractive group drawn from pageant and model settings with a normal group; the direction supports a larger fissure inclination, but the sample is Italian and selected. The 2020 Korean study is a patient sample, so its mean of about 8.4° is not a population norm.

None of the nine gives a cut-off. That is why this page prints degrees and shares, not a pass or fail.

What Moves the Number

The reading is stable only when the photo is. Seven things move it.

Gaze. The pupil line is built from the iris centers, so looking at the screen edge or slightly away from the lens tilts the reference line. Look at the camera.
Head roll, pitch and yaw. The tool rotates the face until the pupil line is horizontal, but residual pose still changes the projection. The quality gate rejects roll beyond 32°, yaw beyond 0.3 and pitch outside 0.10–0.88, and flags roll of 12° or more and yaw of 0.12 or more as low reliability.
Eyelids and eye shape. A heavy lid, a monolid, a visible crease or ptosis changes where the corner points land, even when the corner itself has not moved.
Makeup, lashes and lenses. Eyeliner extends the corner, lashes can cover it, and colored or circle lenses change the iris reading. If you compare two photos, keep the eye condition the same.
Glasses and camera height. Frames can hide a corner, and a camera above or below eye level adds pitch. Keep the lens at eye height and the head straight.
Age. A 2022 study of white adults found the palpebral fissure inclination changes with age, with the inner canthus moving down and outward and the outer canthus down and inward. A single photo cannot separate that from pose.
Expression. A squint or a wide-eyed look changes lid position and can move the detected corners. Use a relaxed, straight-on face.

A good photo: front-facing, eyes open, no heavy eyeliner, no colored contacts, camera at eye level, plain light.

Left and Right Do Not Match — and That Is Ordinary

The median left-right difference is 0.85° in the Western sample and 0.75° in the East Asian sample. At p90 the gap is 2.09° and 2.21°. Half of both samples differ by less than about one degree.

A gap of 2–3° is still inside the ordinary range of these samples. Faces are not mirror images, and a small difference in corner height is common.

The result card calls this number Gap, not asymmetry: the word describes a difference, not a defect.

A sudden change in one eye, a drooping lid, double vision, pain or swelling is a different matter. That needs a clinician, not a photo tool.

Why This Page Has No Score

There is no consensus threshold. The literature reports sample means, directions and significance, not a cut-off above which a tilt counts as good. Population, age and sex all move the ordinary range, so a 0–100 scale would dress up an agreement that does not exist.

The error on one 2D photo is larger than the differences being ranked. Gaze, eyelids, makeup and camera height move the reading; placing +5.1° and +5.8° in different grades claims more precision than the photo supports.

Common is not better. A positive tilt appears in 93.7% of the Western sample and 98.3% of the East Asian sample; a scale that rewards being more positive would crowd most faces near the top and still say nothing about how they look.

If you want the site's index, the AI attractiveness test is that page. It scores two canthal rows; this page reports only the angles.

How This Relates to the Attractiveness Index

The head-page index uses two rows built from these landmarks. The canthal row reads the left-right gap and carries a weight of 3 in both the Western v9 and East Asian v10 profiles. The canthalTilt row compares the mean angle with the performer band and carries a weight of 5 in Western v9 and 7 in East Asian v10, with direction +1.

Inside a band the row score runs from 60 to 100; outside it, the score decays over a tolerance derived from the same sample. That is index calibration, not an aesthetic verdict, and this page shows neither row score.

The readings themselves do not depend on the reference sample: the same photo gives the same left, right, mean and gap under either profile. Only the row scores move.

Open the AI attractiveness test and its seven-row formula.

Frequently Asked Questions

What is canthal tilt?: Canthal tilt is the angle between the line from the inner eye corner to the outer eye corner and the pupil line. Positive means the outer corner sits higher on screen, negative means it sits lower, and the test prints the angle for each eye, their mean and the gap between them. The informal term hunter eyes is not a measurement category here: this page reports one angle, not an eye-shape label.
What does a positive canthal tilt mean?: It means the outer corner of the eye sits higher than the inner corner, on average across the two eyes. In the ordinary samples on this page the majority sit above 0°: 93.7% of the Western sample and 98.3% of the East Asian sample. It is a common shape, not a rare feature, and the label is not a judgement about how a face looks.
Is positive canthal tilt better than negative?: There is no consensus. The studies in the dated table report associations and sample means; none supplies a cut-off, and this page does not score. The head page's canthalTilt row does score against a performer band, but that is index calibration, not an aesthetic conclusion.
Why are my left and right angles different?: The median left-right difference is 0.85° in the Western sample and 0.75° in the East Asian sample, and at p90 the gap is 2.09° and 2.21°. A difference of 2–3° is still inside the ordinary range of these samples. Faces are not mirror images. A sudden change, a drooping lid, double vision, pain or swelling needs a clinician.
Can makeup, eyeliner or contact lenses change the reading?: Yes. Eyeliner extends the corner, lashes can cover it, and colored or circle lenses change the iris position that the pupil line depends on. For a like-for-like comparison, take both photos with the same everyday eye condition, or both with a clean face.
Does head tilt or camera angle change the number?: Yes. The tool aligns the face on the pupil line and rejects photos outside its pose gates, but remaining gaze and pitch still move the reading. Look at the lens, keep the head straight and hold the camera at eye level.
Why is there no score on this page?: Three reasons: the literature has no consensus threshold; the error on a single 2D photo is larger than the differences a rank would separate; and a positive tilt is common, so a higher-is-better scale would not tell faces apart. The AI attractiveness test has an index if that is what you want.
How is this different from the AI attractiveness test?: The head page combines seven proportion rows into a 0–100 index; this page reports only the eye-corner angles. Here there is no total, no rank and no reference switch. Both pages use the same detector and the same 478 landmarks.
Is this a medical diagnosis?: No. This is a 2D landmark angle, not a clinical measurement of the palpebral fissure. Sudden asymmetry, a drooping eyelid, double vision, pain or swelling should be assessed by an eye or plastic-surgery clinician, not by a photo tool.

Method and Sources

Detector: MediaPipe Face Landmarker, 478 points, run locally in the browser. The angle uses eye corners 133, 33, 362 and 263 and iris centers 468 and 473. No image is uploaded and no server-side model runs.

Samples: ordinary stock portraits from Pexels, measured with the same detector. The Western pool processed 1,509 rows and produced 1,301 scored readings; the East Asian pool processed 793 rows and produced 619, of which 534 remain after removing 85 photos that also appear in the Western pool. A face is never counted in both reference bands.

Statistics: population SD (÷ n); quantiles use linear interpolation at (n − 1) × p; readings are kept to four decimals in the tables. Dates in the evidence table are first publication dates from Europe PMC; the source list keeps the journal volume year.

Sensitivity: removing the same 85 photos from the Western pool leaves n = 1,216, moves every band endpoint by 0.08° or less and every share by 0.3 percentage points or less. The page reports the full Western sample and the de-duplicated East Asian sample.

Script and snapshot: scripts/canthal-stats.mjs recomputes these numbers from the local calibration files, which are not public; the checked-in snapshot is docs/ai-attractiveness-test/batches/wave-4-canthal-tilt-test/canthal-samples-2026-10-28.json. The tests read the snapshot, not the photos.

Mean-angle distribution by sample
Sample	n	Mean ± SD	p10	p25	Median	p75	p90	Min	Max
Western, full	1,301	4.9732 ± 3.0860	0.8727	2.9369	5.1013	6.9992	8.9330	−3.9596	13.5352
East Asian, overlap removed	534	6.5545 ± 2.8624	2.7044	4.5432	6.7555	8.5283	10.0670	−3.2337	14.6840
Western, overlap removed	1,216	4.9076 ± 3.0746	0.8394	2.8641	5.0417	6.9427	8.7444	−3.9596	13.5352

Shares and left-right gap by sample
Sample	Above 0°	Above +1°	Within ±1°	Below −1°	Gap median	Gap p90
Western, full	93.70%	89.55%	6.76%	3.69%	0.85°	2.09°
East Asian, overlap removed	98.31%	97.75%	1.50%	0.75%	0.75°	2.21°
Western, overlap removed	93.59%	89.47%	6.74%	3.78%	0.86°	2.11°

Four limits, restated

It is not a population norm. Both sets are convenience samples of stock portraits, so they describe these photos, not Western or East Asian people in general.
It is not a beauty verdict. Positive, neutral and negative are descriptions; inside a band means only that a reading sits in the middle 50% of one sample.
It is not a diagnosis. 2D landmarks move with pose, gaze, eyelids, makeup, glasses, camera and age.
It does not guess ethnicity. Western and East Asian are reference samples you compare with; the page does not infer where you are from.

The photo is processed in the browser and is never uploaded. A reading describes one photo, not a person.

Sources

Bashour M, Geist C. Is medial canthal tilt a powerful cue for facial attractiveness? Ophthalmic Plast Reconstr Surg. 2007;23(1):52–56. PMID 17237692. DOI
Price KM, Gupta PK, Woodward JA, et al. Eyebrow and eyelid dimensions: an anthropometric analysis of African Americans and Caucasians. Plast Reconstr Surg. 2009;124(2):615–623. PMID 19644282. DOI
Vasanthakumar P, Kumar P, Rao M. Anthropometric analysis of palpebral fissure dimensions and its position in South Indian ethnic adults. Oman Med J. 2013;28(1):26–32. PMID 23386941. DOI
Sforza C, Dolci C, Grandi G, et al. Comparison of soft-tissue orbital morphometry in attractive and normal Italian subjects. Angle Orthod. 2015;85(1):127–133. PMID 24857159. DOI
Lee JY, Choung HW, Choung PH. Diagnostic analysis of vertical orbital dystopia and canthal tilt for surgical correction. J Korean Assoc Oral Maxillofac Surg. 2020;46(6):379–384. PMID 33377462. DOI
Liu J, Rokohl AC, Liu H, et al. Age-related changes of the periocular morphology: a two- and three-dimensional anthropometry study in Caucasians. Graefes Arch Clin Exp Ophthalmol. 2023;261(1):213–222. First published 2022-08-15. PMID 35969325. DOI
Liu AS, Salinas CA, Sharaf BA. Using artificial intelligence to quantify sexual dimorphism in aesthetic faces: analysis of 100 facial points in 42 Caucasian celebrities. Aesthet Surg J Open Forum. 2023;5:ojad046. PMID 37441561. DOI
Gao T, Guo Y, Rokohl AC, et al. Racial and sexual differences of eyebrow and eyelid morphology: three-dimensional analysis in young Caucasian and Chinese populations. Quant Imaging Med Surg. 2025;15(1):882–897. First published 2024-12-30. PMID 39839011. DOI
Wang B, Ran Z, Deng Y, et al. Associations between palpebral fissure morphology and AI-derived facial attractiveness scores in Asian adults. J Plast Reconstr Aesthet Surg. 2026;123:135–143. PMID 42863361. DOI