Skip to main content
AI Rate Face
Local feature tool

Eye Shape Detector

Get a conservative eye-geometry label plus width-to-opening ratios, tilt and symmetry. The tool avoids claims it cannot reliably infer from landmarks alone.

Analyze a photo locally →
On-device analysis

Estimate visible eye geometry

Look straight at the camera with a relaxed expression and unobstructed eyes. Analysis stays on your device.

Photo stays in this browser

Choose one clear adult portrait

Front-facing, neutral expression, even light. JPG, PNG, WebP, HEIC or HEIF up to 10 MB.

or drag and drop it here

Before you analyze

  • 1Use your own photo, or one you have permission to use.
  • 2One adult face only. No minors or age estimation.
  • 3Results describe this photo, not objective beauty or health.

A typical first run loads about 15 MB of model/runtime files, then the browser caches them. Nothing is submitted to our analysis API.

Local mode: the selected image and detected landmarks are held in page memory only and are cleared when you reset or leave.

Drag to reposition
Local face crop

Adjust the face area

Keep one full face inside the oval. Leave the hairline and chin visible, with a little space around the head.

For more reliable detection

  • • Select only one adult when the photo contains several people.
  • • Avoid cutting through the forehead, chin or either side of the face.
  • • Cropping helps framing, but cannot restore blur or very low resolution.

The crop is created in this browser and is never uploaded.

What it measures

Evidence behind this result

The output is built from inspectable photo geometry and a separate image-quality signal.

01

Width / opening

Compares horizontal eye width with average vertical lid opening in the photo.

02

Outer-corner tilt

Measures the visible outer-to-inner corner angle for each eye.

03

Side agreement

Combines visible size, opening and angle differences.

01Visible landmarks
02Normalized geometry
03Explainable output
Three-step method

How this tool works

01

Outline

Six landmarks around each visible eye define its simplified geometry.

02

Compare

Width-to-opening ratios produce a conservative shape-family label.

03

Qualify

Pose and photo-quality signals accompany the result.

478 visible landmarksMediaPipe Tasks Vision 0.10.35Model: float16 v1Runs in browser
Read the shared methodology →
Interpret carefully

What the result cannot tell you

01

The tool cannot reliably classify monolid, hooding or deep-set anatomy from sparse 2D landmarks.

02

Squinting, smiling, makeup, glasses and eyelashes can change the visible opening.

03

The labels describe geometry in one photo and are not ethnic or biological classifications.

Questions, answered

About this tool

How local processing works →
Why are there only three eye-shape labels?+

The landmark model supports conservative width-and-opening geometry. More specific labels would require visual tissue interpretation that this tool does not claim.

Should I remove glasses?+

Yes if comfortable. Clear, unobstructed eye corners and lids improve the estimate.

Does smiling change the result?+

It can. A smile may narrow the visible lid opening, so a relaxed expression is best for repeatability.

Keep exploring

Related tools

View all tools →
Adult portrait with a complete face-analysis landmark overlay
AI report

AI Face Rater

A free on-device face score, face-shape match, symmetry breakdown and practical photo and styling plan.

Open tool →
The same adult portrait shown under two different lighting conditions
Photo decisions

Compare Two Photos

Upload two photos and see which one presents better, broken down by lighting, sharpness, head angle and framing.

Open tool →
Three aligned photos of the same adult face for a consistency comparison
Photo decisions

Face Score Consistency Test

Run two or three photos of the same face to see how much of a score change is the face and how much is the photography.

Open tool →