Opening
How wide the visible opening is compared with its central height. Round, almond and elongated describe different opening contours.
What eye shape do I have? Measure your eye opening, corner tilt and spacing from one photo. Add an optional eyelid self-check for styling ideas.
Free · No signup · Your photo stays on your device
Drag and drop or choose a file
JPG, PNG up to 10MBFace the camera at eye level. Keep both eyes naturally open and remove glasses.
Explore your eyes through three automatic photo measurements—opening shape, corner tilt and spacing—and an optional eyelid self-check. These features can overlap: an almond-like opening can also have upturned corners and a covering lid fold.
How wide the visible opening is compared with its central height. Round, almond and elongated describe different opening contours.
Whether the outer corner sits below, level with or above the inner corner: downturned, neutral or upturned eyes. This angle is called canthal tilt.
Eye spacing compares the inner-corner gap with eye width. Close-set and wide-set describe relatively smaller and larger gaps; this tool shows the ratio rather than assigning either label.
Crease visibility and skin coverage are separate. Hooded describes a covering fold; monolid refers to an eyelid without a distinct crease. A hidden crease alone does not establish either.
Schematic illustrations. Crease visibility and skin coverage can overlap; the tool does not automatically classify either.
The model finds facial landmarks on your device. Geometry measures three features from those points.
Eye width runs from the inner to the outer corner. Height uses the central upper and lower lid points, projected perpendicular to that corner line. Each eye gets its own width-to-height ratio, so small differences remain visible.
This tool’s guide bands call ratios below 2.6 round-like, from 2.6 to 3.3 almond-like, and above 3.3 elongated. A 0.1 buffer around each boundary shows both descriptions. These are product guide bands, not clinical norms or calibrated confidence intervals. A naturally narrow opening is not evidence of a closed eye.
The tool corrects minor sideways roll using a line between landmark-based eye centers, then measures each inner-to-outer corner angle. Positive means the outer corner sits higher in this photo. The −1° to +1° neutral range matches our Canthal Tilt Test; it is a display convention, not a healthy or ideal range.
The distance between the two inner corners is divided by the mean width of your eyes. A reading of 1.20× means the gap is 1.2 times that average width. One eye width is a comparison reference, not a standard of attractiveness. This is different from pupil distance and cannot be used to order prescription glasses.
Landmarks trace eye edges, but do not locate the upper-lid crease or covering skin fold. You can describe those yourself after the scan. A hidden crease does not automatically mean monolid, and hooding is separate from whether an opening looks round or almond-shaped.
The detector checks face count, image and eye size, light, fitted head position and eyelid geometry. It withholds readings when those checks fail. Passing them does not prove the photograph is free of perspective, squinting or makeup effects. It does not detect glasses, diagnose eyelids, measure orbital depth, infer personality or rank attractiveness.
Opening labels and photo-quality thresholds are product conventions, not clinical standards. Head-pose checks are conservative; a rejected photo is a request for a clearer view, not a judgment about your eyes.
Keep the lens at eye level and your chin relaxed. Avoid side views or looking up at the camera.
A squint, smile or deliberately widened gaze changes the opening. Use a relaxed expression.
Remove glasses and move hair away from the corners. Strong liner, lash extensions and glare can alter the outline.
Face a window without harsh brow shadows. Choose a sharp portrait rather than a distant face or an eye-only crop.
Use a well-lit mirror and relaxed eyes. This is a way to describe what you see, not a diagnosis or another AI scan.
Compare its width with its central height. Is it relatively open and round, or longer and narrower? You may sit between descriptions.
Keep your head level. Notice whether the outer corner appears above or below the inner one. Check both eyes separately.
Compare the inner-corner gap with the width of one eye. Think of it as a proportion, not a target to meet.
Can you see a crease with your eyes naturally open? How much of the moving lid above your upper lashes is covered by skin above it? These are separate observations, and “not sure” is a useful answer.
Start with one clear, front-facing portrait and naturally open eyes. The detector measures the visible opening, corner tilt and space between your eyes. Check the drawn lines against your photo before using the description.
Yes. Almond describes the opening, while hooded describes skin covering part of the upper lid. Opening, tilt, spacing and lid appearance are different observations, so more than one description can fit.
No. The face mesh estimates lid edges and corners, not the upper-lid crease or skin fold. You can optionally describe crease visibility and lid coverage yourself. Your choices are marked as your selection and do not change the measured numbers.
Squinting, expression, gaze, camera distance, lighting and head position can change the photographed opening. Use a relaxed expression and the same eye-level setup when comparing pictures. Roll correction cannot remove three-dimensional perspective.
Yes, and a photo can make a small difference more visible. The tool keeps both sides separate instead of forcing their average into one label. Left and right refer to the sides of the displayed photo.
These are two-dimensional photo estimates, not clinical measurements. The opening descriptions use product guide bands, not population norms. Near a boundary the tool shows both descriptions. A decimal place is display precision, not a guarantee of accuracy. The tool cannot reliably distinguish every naturally narrow eye from a squint in a single image.
No. The face model downloads to your browser and the photo is analyzed there. Photos, landmarks, results and eyelid selections are not uploaded or saved in browser storage. Save result creates a local PNG. Clear photo or refresh to remove the current result. Loading the page and model still requires network requests.
The tool is free and needs no account. If you prefer not to use a photo, try the mirror self-check above. It is a visual guide, not an automated scan.