How the eye shape detector works
- 1
Add a front-facing photo
Relax your eyes and look straight into the lens. Before measuring, a built-in quality check rejects photos with no face, more than one face, a face that is too small or too dark, or a head turned or tilted too far.
- 2
Points around each eye
A face landmarker model runs directly in your browser tab and estimates 478 3D points across your face, including the lid lines, the inner and outer eye corners, and a ring of points above each eye. Your photo stays on your device and is never uploaded.
- 3
Measure, then name
Your canthal tilt and eye height to width decide the shape word in a fixed order. Eye spacing and upper lid exposure are shown alongside it, with the measurement lines drawn on your photo so you can see how each number was read.
What we measure
Eye shape names are loose everyday words. We tie each label to concrete measurements on the face mesh, so the same photo always gives the same answer.
Eye height to width
DeviceThe vertical opening between the upper and lower lid lines divided by the horizontal corner-to-corner width of the eye. A higher number means a taller, more open eye, while a lower number points to a longer, narrower opening.
Canthal tilt
DeviceThe angle of the line from the inner eye corner to the outer eye corner, averaged over both eyes. A positive angle means the outer corner sits higher than the inner corner.
Eye spacing
DeviceThe gap between the two inner eye corners divided by the average width of your eyes. A value near 1.0 means the space between your eyes equals one eye width, and higher numbers mean wider-set eyes.
Upper lid exposure
DeviceThe distance from the upper lid line to the crease ring of points above the eye, divided by eye width. Your result is grouped into low (below 0.205), medium (0.205 to 0.245) or high (above 0.245).
Canthal tilt: why slant comes first
Canthal tilt describes whether the outer corners of your eyes angle upward, sit nearly level, or angle downward relative to your inner corners. People who talk about upturned or downturned eyes are describing this slant. Our detector checks canthal tilt first, before looking at how tall or narrow the eye opening is.
To measure canthal tilt in your browser, the 3D face mesh locates the inner and outer corners of both eyes. The tool measures the angle of the line from each inner corner to its outer corner, then averages your left and right angles into a single mean canthal tilt.
Positive tilt
Level corners
Negative tilt
Each eye is drawn from inner corner (left) to outer corner (right). Pink: corner line; dashed: face horizontal.
Eyelid geometry is a subject of published research. Rhee, Woo, and Kwon (2012) took photogrammetric measurements of composite faces, including the slant of the palpebral fissure, and reported differences between composites rated attractive and average ones. Kunjur, Sabesan, and Ilankovan (2006) measured the position of the eyebrow and eyelids in three groups to quantify how these landmarks differ. These studies describe the groups they measured; they do not set a target for your own eyes.23
In our engine, a mean canthal tilt above +7.7° classifies your eyes as upturned, and a tilt below +1.3° classifies them as downturned. Anything between +1.3° and +7.7° moves on to the height-to-width check. The mesh is fitted to a canonical face, so real faces span a narrow band of values. On our calibration faces (133 real front-facing photos, used as a mesh reference and not as population norms), the middle 80% of left-eye canthal tilt ranged from 1.0° to 8.7°, with a median of 4.8°. Brow shape and a camera held above or below eye level can change how steep the line looks in a photo, so use an eye-level, straight-on shot.
Eye height to width: almond, round, or narrow
Once a photo falls in the middle range of canthal tilt, eye height to width separates round, narrow and almond eyes. Putting the proportion into a single ratio removes guesswork, especially when an eye sits close to the border between two categories.
The face landmarker places points along both the upper and lower eyelid margins as well as the inner and outer corners. The tool compares the vertical gap between the lid lines with the horizontal corner-to-corner width. Because the result is a ratio, it does not depend on how large your face is in the frame. The mesh approach builds on published work by Kartynnik and colleagues (2019), who describe inferring an approximate 3D face mesh from a single camera image.1
Almond
Round
Upturned
Downturned
Narrow
Number of photos
- Median:
- 0.294
- Middle 80%:
- 0.207–0.357
- 1Narrow < 0.235
- 2Round > 0.35
Eyelid researchers also measure fissure height and width. Rhee, Woo, and Kwon (2012) took fourteen distance and five angle measurements on seven composite faces and reported differences in fissure height and width between attractive and average composites, with the direction depending on the group studied.2
If neither tilt rule applies, an eye height to width ratio above 0.35 is labeled round, a ratio below 0.235 is labeled narrow, and anything from 0.235 to 0.35 is labeled almond. On our calibration faces, the middle 80% of eye height to width ranged from 0.207 to 0.357 with a median of 0.294, and 73 of the 133 calibration photos were labeled almond. A squint or a wide-eyed expression can move the upper and lower lid points and push an eye toward narrow or round. Relaxing your eyelids while looking straight at the lens gives the cleanest baseline.
Eye spacing: measuring the gap between inner corners
Eye spacing tells you how far apart your eyes sit relative to their own width. It does not change your eye shape label, but seeing it next to your shape shows why two people with the same label can have different proportions around the eyes.
To compute eye spacing, the detector measures the distance between your left and right inner corners on the 3D mesh. It then divides that inner-corner gap by the average width of your left and right eyes. Expressing the gap in eye-width units removes the effect of photo scale.

Pink dots mark detected facial features; pink lines connect the measured widths, face length and eye corners.
Rhee, Woo, and Kwon (2012) reported that, across the composite faces they studied, those rated attractive had wide-set eyes and a lower brow position than average composites. This is a finding about composites, not a goal for your own photo.2
On our calibration faces, the middle 80% of eye spacing ranged from 1.15 to 1.42 eye widths, with a median of 1.28. Because the mesh is fitted to a canonical face, real faces span a narrow band, so compare your number with these bands rather than with a ruler. Brow grooming and glasses frames can change how the space between your eyes looks, but the measured number only reflects the landmark positions.
Upper lid exposure: low, medium, and high lid show
Upper lid exposure measures how much vertical space sits between your upper lid line and the crease ring of mesh points above your eye when you look straight ahead. It adds detail that the broad eye shape name does not capture.
Because the 478-point face landmarker tracks surface geometry rather than skin folds directly, our tool estimates upper lid exposure from two sets of mesh points. It measures the distance from the upper lid line to the crease ring above each eye, then divides it by your eye width so the ratio holds across different photo sizes.
Upper eyelid position is also studied in eyelid research. Kunjur, Sabesan, and Ilankovan (2006) measured eyebrow and upper eyelid position in three groups to quantify how these landmarks differ, and Rhee, Woo, and Kwon (2012) included double fold line measurements in their composite-face study.23
Our detector groups upper lid exposure into three tiers set from the tertiles of our calibration faces: low is below 0.205, medium is 0.205 to 0.245, and high is above 0.245. These tiers describe our calibration photos, not population norms. Head pitch can change the measured exposure, which is one reason the quality check rejects photos where the head is tilted too far.
The five eye shapes
The shape is decided in this exact order. Thresholds apply to mesh measurements, not physical ruler values, and any bands quoted from our calibration photos are not population norms.
| Shape | Rule | What it looks like |
|---|---|---|
| Upturned | Tilt above +7.7° | Outer corners lifted |
| Downturned | Tilt below +1.3° | Little or no lift at the outer corners |
| Round | Height to width above 0.35 | A taller, more open eye |
| Narrow | Height to width below 0.235 | A longer, lower eye opening |
| Almond | Everything in between | Most common on our calibration faces (73 of 133) |
What the eye shape detector can't tell you
- It cannot detect a monolid or a hooded crease directly; upper lid exposure is estimated from the ring of mesh points above the upper lid line.
- The 3D mesh is fitted to a canonical face, so mesh ratios sit in a narrower band than physical tape-measure or caliper values.
- Glasses frames, lens reflections, heavy eyeliner, and false lashes can shift where the lid and corner points land.
- One photo captures one expression and head angle; smiling, squinting, or tilting your chin up or down can change the measured opening and tilt.
- These measurements describe geometry only. They are not a medical assessment, an ethnicity classifier, or an attractiveness score.
Photo tips
Face the camera straight on
Hold the camera at eye level and look directly into the lens. Turning your head left or right or tilting your chin up or down can distort corner angles and spacing, and the quality check rejects photos turned or tilted too far.
Even, soft light
Face a window or a soft front light so both eyes are lit evenly. A photo that is too dark fails the quality check, and harsh side shadows can hide the inner or outer eye corners.
Neutral expression
Relax your face, forehead, and eyelids before taking the shot. Smiling or raising your brows can change the lid opening and the upper lid measurement.
Hair and glasses off the face
Brush bangs or loose strands away from your eyes and brows, and remove glasses if you can. Clear visibility of both eye corners and the lid lines helps the landmarks land where they should.
Questions
References
- Kartynnik, Y., Ablavatski, A., Grishchenko, I., & Grundmann, M. (2019). Real-time facial surface geometry from monocular video on mobile GPUs. arXiv:1907.06724. arxiv.org/abs/1907.06724
- Rhee, S. C., Woo, K.-S., & Kwon, B. (2012). Biometric study of eyelid shape and dimensions of different races with references to beauty. Aesthetic Plastic Surgery, 36(5), 1236–1245. doi.org/10.1007/s00266-012-9937-7
- Kunjur, J., Sabesan, T., & Ilankovan, V. (2006). Anthropometric analysis of eyebrows and eyelids: An inter-racial study. British Journal of Oral and Maxillofacial Surgery, 44(2), 89–93. doi.org/10.1016/j.bjoms.2005.03.020