Width
Measured straight across, from the inner corner to the outer corner.
Free AI Tool — Ratio-Based
Eye Shape Detector
An eye shape detector is an AI tool that measures the width-to-height ratio and corner angle of each eye from facial landmark points, then classifies the shape as almond, round, upturned, downturned, or somewhere in the middle.
Most people have never measured their own eyes — they've only ever guessed from a mirror or copied whatever a makeup tutorial told them to assume. Upload one photo and this tool draws the outline itself, then hands you the ratio it used to get there.
12 points
per pair
per pair
Never stored,
always private
always private
2 free scans
daily, no sign-up
daily, no sign-up
1 Upload Your Photo
Reading your eyes…
AI analysis in progress
2 Your Eye Shape
Your eye shape, the ratio behind it and your own eye outlines will appear here.
Share your result
How Your Eye Shape Is Read
Twelve points do the work here — six placed around each eye, drawn from the same underlying 68-point facial landmark map used across the rest of the site. From those twelve points, two separate readings get pulled: how wide each eye is relative to its height, and whether the outer corner sits above or below the inner one.
Height
Averaged from a top point and a bottom point, so a mid-blink frame doesn't quietly wreck the number.
The ratio
Divide width by height. Above 3.0 reads almond. Below 2.0 reads round.
Corner tilt
When the ratio lands in between, the angle of the outer corner against the inner one decides upturned versus downturned.
Eye shape happens to be one of the rare facial traits where one division problem carries most of the answer. Width over height already separates almond from round before the model looks at anything else — which is exactly why that figure gets checked first, and corner angle only gets pulled in to settle the middle ground.
Both eyes get reported on their own, not just averaged together. When the two numbers land close, that's expected — most faces aren't perfectly symmetrical. When they land far apart, it's usually telling you something about the photo, not your face, and the next section explains exactly how far apart is worth worrying about.
What the Ratio Means
Width divided by height, averaged across both eyes. None of these three bands ranks above another — they're categories, not a scoreboard.
Above 3.0
Almond
Clearly longer than it is tall, tapering to a point at both corners.
2.0 – 3.0
Middle ground
The ratio alone doesn't decide it here. Corner angle takes over and sorts the result into neutral, upturned, or downturned.
Below 2.0
Round
Height sits close to width, which means more iris stays visible even at rest.
Getting an Accurate
Eye Reading
Height is the single most breakable number this tool produces — a fraction of a second either way changes it.
Eyes fully open
A half-blink shrinks the height reading and can quietly flip a round result to almond.
No glasses
Frames and lens curvature both nudge where the corner points actually land.
Square to the camera
Turn your head even slightly and the far eye foreshortens, which is usually why the two sides disagree.
Even lighting
Deep shadow pooling in the eye socket can hide the upper lid points the model needs.
How the Detector Works
1
Upload Photo
One clear, front-facing shot with both eyes open, nothing fancier required.
2
Points Located
Six landmarks get placed around each eye, twelve total.
3
Ratio Measured
Width divided by averaged height, calculated per eye.
4
Shape Named
Ratio decides first; corner angle only steps in if the ratio sits in the middle band.
The same landmark logic runs underneath the site's nose shape detector — different points, different ratios, same underlying idea of letting geometry answer a question a mirror can't.
Why Choose
FaceShapesAI?
Advanced AI technology combined with beauty science to give you the most accurate results.
Advanced AI Technology
Trained on millions of faces for high accuracy.
Privacy Focused
Your photos are never stored. 100% secure & private.
Instant Results
Get your analysis in seconds, not minutes.
Personalized Tips
Receive style, haircut & grooming recommendations.
User Friendly
Simple, easy-to-use interface for everyone.
Works for Everyone
Our AI works for all genders, ages, and ethnicities.
Why Eye Shape Lists Disagree With Each Other
Go looking for "eye shapes" and the count changes depending on where you look — six types on one list, ten on another, and barely any overlap between them. Some throw in hooded and monolid. Some treat deep-set and protruding as their own categories. Some collapse almond and upturned into a single entry.
The disagreement makes sense once you notice those lists are mixing several different kinds of description into one bucket. Almond versus round is a proportion question. Upturned versus downturned is an angle question. Hooded and monolid describe the eyelid's fold structure — a completely separate axis, and one a flat outline of the eye opening isn't built to see at all.
This tool sticks to the two axes it can measure with confidence: proportion and corner angle. It doesn't claim a verdict on hooded or monolid lids, because the six points per eye trace where the eye opens, not what's happening in the skin above it.
What the Numbers Actually Show
Reporting both eyes separately isn't just for show — it's diagnostic. Most people's eyes are close but not identical, and that's normal. The number worth paying attention to is the gap: when the two ratios split by more than roughly 0.4, the cause is almost never your actual face. It's a head turned slightly off center, which foreshortens the farther eye, shrinks its measured width, and drags its ratio down with it.
Treat that gap as a built-in photo check rather than a curiosity. A wide split between the two numbers is a cue to retake the shot square-on before trusting the label attached to it.
About the Eye Colour Reading
Colour comes from an entirely different calculation — the dominant hue sampled from the pixels inside the iris outline, nothing to do with the twelve shape-landmarks above. It's also the most lighting-sensitive reading anywhere on the site: warm indoor bulbs push hazel toward amber, while a cool monitor glow or an overcast sky can drag brown toward grey.
A result that looks off is worth re-running near a window in daylight before assuming it's wrong. Very dark irises are a harder case regardless of lighting — there's so little saturated colour to sample that the reading defaults to brown rather than guessing at a hue it can't see.
About This Tool
An in-house model reads the ratio above, not a licensed third-party API — Ahsan Fayyaz trained it on real faces and benchmarked it against other face-analysis tools before it went live. It's the same landmark system running the nose and lip reading tools elsewhere on the site. Results are for informational and entertainment purposes only, not medical, clinical, or professional advice, and shouldn't be treated as a diagnosis of anything. Browse the rest of the tools directory for the other face-measurement readings this same engine powers.
Our Growing Suite of AI Face & Style Tools
Explore our powerful AI tools designed to enhance your style and confidence.
Face Shape Detector
Discover your face shape with AI accuracy.
68-Point Landmark Map
See every point the AI reads on your face.
Eye Shape Detector
Almond, round, upturned or downturned.
Nose Shape Detector
Base width against bridge length, from the front.
Lip Shape Analyzer
Fullness, bow shape and proportion.
Face Symmetry Test
A region-by-region symmetry breakdown.
Symmetry Mirror & Heatmap
Your face mirrored on both sides.
Golden Ratio Proportions
Three facial ratios measured against phi.
Golden Ratio Phi-Mask
The phi grid drawn over your photo.
Celebrity Look-Alike
Find the celebrity your face matches best.
K-pop Look-Alike
See which K-pop idol you resemble.
Hairstyle Recommender
Cuts matched to your face shape.
Beard Style Finder
Beard shapes that suit your jaw.
Glasses Frame Finder
Frames that flatter your proportions.
Sunglasses Shape Guide
The right sunglasses for your face.
Earring Shape Guide
Earrings that balance your face shape.
Neckline Guide
Necklines that suit your face and frame.
Hat & Cap Guide
Hat shapes that work with your features.
Fringe Suitability
Should you get bangs? Find out first.
Hair Parting Guide
Middle, side or deep side parting.
Makeup Contour Guide
Where to contour for your face shape.
Eyebrow Shaping Guide
The brow shape that fits your features.
Beard Length Advisor
How much length balances your jaw.
Skin Tone Analyzer
Your skin depth across five bands.
Body Shape Calculator
Work out your body shape from measurements.
Body Ratio Calculator
Waist-to-hip and shoulder ratios.
Outfit Silhouette Guide
Silhouettes that flatter your shape.
Dress Style Recommender
Dress cuts matched to your proportions.
Universal Size Helper
Translate your measurements across brands.
Got Questions?
Frequently Asked Questions
It places six landmark points around each eye — twelve total — then divides width by height to get a ratio. Above 3.0 lands as almond, below 2.0 as round, and anything in between gets sorted by corner angle into neutral, upturned, or downturned.
Because a single number is harder to argue with than an impression. Two eyes that look similar at a glance can carry meaningfully different ratios once they're measured, and the model only trusts the math.
No, and it won't pretend to. Those describe the eyelid's fold structure, which sits on a different axis entirely from proportion and corner angle — the six points per eye trace the opening itself, not the skin above it.
Usually, yes — most faces aren't perfectly symmetrical, so a small gap is expected. It's only worth a second look once the two ratios split by more than about 0.4, since that size of gap is almost always a slightly turned head rather than an actual asymmetry.
They can. Frames sit close enough to the eye that both the rim and any lens curvature can shift where the corner points get placed, which nudges the ratio either direction.
No. It's processed by the in-house model and deleted from the server right after your result comes back. What you see afterward lives only in your browser's cache — refresh the page and it's gone for good, on both ends.
Two, no account required.
It raises your daily scan limit above two. Paid plans remove the daily cap entirely.
They come from two different calculations. Shape uses the twelve landmark points; colour samples the dominant hue from the pixels inside the iris. One reads geometry, the other reads pixel colour, and they can be affected by completely different things — a filter might shift colour without touching shape at all.
Almost always lighting. Warm indoor bulbs push hazel toward amber, and a cool screen glow or overcast sky can drag brown toward grey. Try again near a window in daylight before assuming the reading is broken.
Very dark irises carry little saturated colour for the model to sample, so the reading defaults to brown rather than guessing at a shade it can't clearly see.
It can. A big smile pulls the cheeks up and narrows the eye opening more than most people expect, which can shrink the height reading and shift the ratio toward almond even though your resting eye shape hasn't moved.
Front-facing, both eyes open, no glasses, even lighting, and your head square to the camera rather than turned. Each of those affects a different part of the measurement, so skipping more than one tends to compound the error.
No — the same image produces the same landmark points and the same ratio every time. A different result only shows up with a different photo, angle, or expression.
Rarely for shape, since the model is finding the edge of the eye opening itself rather than reading makeup lines. Colour is more sensitive — heavy shadow or a colored lens can shift the sampled hue.
The twelve eye points are drawn from that same underlying map, yes — just the subset relevant to the eyes rather than the full face. The landmark map page shows how all 68 points are laid out together.
It can, gradually, since skin and lid position shift over decades and that can nudge where the corner points sit. The tool only reads what's in the photo you give it, so a current photo reflects your current geometry, not a fixed trait from years ago.
Face shape reads the jawline, cheekbones, and forehead to classify the outline of the whole face. This tool is narrower on purpose — it only looks at the eyes, using the same landmark approach the site's nose shape and lip tools apply to their own regions.
Yes — your result, ratio, and outline can be posted straight to social media as a styled image, no screenshot needed.
Ready to Discover Your Look?
Join thousands of people already using our free AI tools to understand their features and style.
Try It Free Now
No sign up required
Ready to Find Your Eye Shape?
Upload a photo and see your ratio, your corner angle, and the shape they add up to.