INPUT
One clear front view
Unusable images are rejected instead of receiving a guessed score.
Measure the tilt from inner to outer eye corner, separately for each eye. Get signed degrees and a diagram of the points used, with camera roll removed.
JPG, PNG or WebP · One face, straight on, in even light.
Processed in this browser. Your photo stays in this tab until you clear it or close the tab.
Canthal tilt is the angle from the inner corner of the eye to the outer corner. This test measures it after removing camera roll, reports it separately for the left and right eye, and states the sign convention so the number can be read without guessing. It is a standalone measurement and is not part of the 0–100 composite.

INPUT
One clear front view
Unusable images are rejected instead of receiving a guessed score.
OUTPUT
Left and right tilt, in degrees
Positive means the outer corner is higher; negative means it is lower. Eye tilt is reported separately from the composite score.
01
Four landmarks are used: the inner and outer corner of each eye. The engine rotates the face upright first, then takes the angle between the inner-to-outer vector and the horizontal, using atan2 for a signed result in degrees. Rotating first is the whole point — without it, a head tilted five degrees reports five degrees of canthal tilt that the face does not have.
Sideways tilt is the only head movement the engine can undo. A head turned left or right is rejected by the pose gate. A head tipped forward or back is neither corrected nor rejected, and it changes what the camera sees: in a study that photographed 12 volunteers at head angles from −15° to +15°, tipping the head down made the line between the eye corners look more upturned (Vaca et al., 2021). Keep the camera at eye height and your chin level, or part of the number is your posture.
02
The test reports a signed number. The bands below put words to it. They describe the slope of a line — nothing more. They are not population norms, not a diagnosis and not a beauty ranking, and a higher number is not a better number. The middle band is ±2° wide because that is the tolerance this test already uses for its reference band: a reading inside it is level as far as this measurement can tell.
| Measured angle | Label | What the line does |
|---|---|---|
| Above +8° | Strongly upturned | The outer corner sits well above the inner corner. |
| +4° to +8° | Upturned | A clear upward slope toward the temple. |
| +2° to +4° | Slightly upturned | A small upward slope. |
| −2° to +2° | Level | Within the ±2° tolerance of horizontal. |
| −4° to −2° | Slightly downturned | The outer corner sits a little below the inner corner. |
| Below −4° | Downturned | A clear downward slope toward the temple. |
A value on a boundary belongs to the band closer to zero, so +4.0° reads as slightly upturned. The scan result itself shows the signed number and the reference band; these labels are for reading it.
03
The sign convention is stated rather than implied: positive means the outer corner sits higher than the inner corner, for both eyes. Negative means the outer corner sits lower. Left and right are reported separately, because they are frequently not the same number and averaging them would hide that.
Online, a positive tilt is often called "hunter eyes" and a negative one "puppy eyes" or "droopy eyes". Those words describe a whole eye shape — how much lid shows, where the brow sits, how far apart the eyes are. The test measures one angle and does not use either label. For a separate measurement of how far apart the iris centres sit relative to face width, see the pupil spacing ratio.

04
Less than the internet suggests. A 2024 review of the evidence behind facial aesthetic ideals found that many published standards have little empirical analysis behind them; the eye region was covered by six studies (Faderani et al., 2024).
The largest direct test points the other way from the online consensus. 2,081 people rated an edited face in which eyebrow height and outer eye-corner height were changed. Raising the outer corners lowered ratings of attractiveness and trustworthiness and raised ratings of dominance and threat (Kempa et al., 2025). That is one face in one survey, but it is a reason not to read a higher tilt as a better one.
Typical values depend on who is measured and how. In a clinical study of 76 Korean patients, the mean canthal tilt was about 8.4° on each side (Lee, Choung and Choung, 2020). One clinical sample from one population is not a norm for everyone, and this test has not been calibrated against it.
05
Because faces are not mirror images. In the same clinical study, the outer eye corners sat at different heights by about 2 mm on average (Lee, Choung and Choung, 2020). Across an eye roughly 30 mm wide, that alone is a few degrees.
The photo adds its own differences. A slight head turn moves the near and far eye corners by different amounts, a squint or a smile pulls the outer corners, and a lens held close to the face exaggerates whatever is nearest. The test reports each eye separately rather than averaging, so you see the gap instead of having it hidden.
06
A canthal tilt test is only as steady as the photo. Hold the phone at eye height, about an arm’s length away. Keep your chin level — not tucked, not raised — because that is the one movement this test cannot correct. Relax your face: no smile, no squint. Use even light from the front and take glasses off.
Then scan two or three times. If the readings move a lot between scans, the setup is changing, not your face.
07
The canonical face measures 1.74 degrees on both eyes, and the reference band is that value plus or minus 2 degrees. That tolerance is a fixed engineering tolerance chosen for the measurement, not a population norm — nobody has surveyed a population here, and the page will not imply one.
08
This is a front-view angle in the image plane. A turned head projects the eye corners onto different positions and is rejected by the pose gate rather than measured. Eyelid shape, the position of the eye within the socket, and anything requiring depth are not measured at all.
09
Canthal tilt does not feed the 0–100 composite. That number is built from exactly four weighted dimensions — symmetry 35%, facial thirds 25%, facial fifths 25%, reference ratios 15% — and adding a fifth input would break a promise printed on every page of this site. Canthal tilt is reported next to the score, not inside it.
10
The eye-corner coordinates are computed in your browser and stay there. No photo, no landmark, no angle, and no record of the measurement reaches our servers. There is no account and no payment.
FAQ
It is the angle of the line from the inner corner of the eye to the outer corner, relative to horizontal. A positive tilt means the outer corner sits higher; a negative tilt means it sits lower.
From four landmarks — the inner and outer corner of each eye — after the face has been rotated upright and rescaled. The angle is computed with atan2 and reported in degrees, separately per eye.
1.74 degrees on both eyes, with a fixed reference band of plus or minus 2 degrees around it. That band is an engineering tolerance, not a statement about what is common or desirable.
Because they often are, and the measurement reports each eye rather than averaging. The ±2° band shown with each value is a fixed engineering tolerance chosen for repeat measurement and display, not a population norm and not a clinical threshold.
No. It is a direction of slope, not a defect. Research on how tilt affects ratings is limited, and the largest rating study found that raising the outer eye corners lowered attractiveness ratings rather than raising them. This test reports the angle and attaches no verdict to it.
There is no agreed normal value. Published averages come from specific clinical samples — one study of 76 Korean patients reported about 8.4° — and they depend on the population and on how the angle was measured. This test does not compare you with anyone; it reports your angle next to a fixed ±2° band around its reference face.
Yes. It finds the inner and outer corner of each eye automatically and calculates the angle in degrees for each eye. You do not place points or measure anything by hand.
Tilting sideways does not: camera roll is removed before the angle is taken. Tipping your head forward or back is not corrected, and research on head position found that tipping the head down makes the eye corners look more upturned. Keep your chin level.
No. The 0–100 composite uses four weighted dimensions and canthal tilt is not one of them. It is measured and displayed alongside the score, never folded into it.
No. A turned head is rejected by the pose gate before any angle is computed, because projection moves the eye corners in a way the front-view measurement cannot undo.
Run the free scan in your browser to see left and right tilt, in degrees, the measurement points and the reference values.