Free · Photo, camera or measurements
Face shape calculator
Find the face shape your proportions are closest to. Enter five measurements, upload a photo, or use your camera, then see the ratios behind the result and how close the runner-up was.
- 3
- ways to test
- 6
- ratios compared
- 10
- shapes scored
- 478
- landmarks in the model
- Runs in your browserPhotos are analyzed on your device
- No accountNothing to sign up for
- Every score shownNot just one label
- Plain about accuracyHeuristic profiles, not a diagnosis
What is a face shape calculator?
It estimates which named shape your proportions are closest to, and shows the numbers behind the answer.
The definition
A face shape calculator estimates which named shape (oval, round, square, heart, diamond, oblong, rectangle, triangle, pear or inverted triangle) your face is closest to. It does this by comparing proportions: face length to cheekbone width, forehead to cheekbone, jaw to cheekbone, chin to jaw, and jaw angle.
How to use it
Enter five measurements by hand, upload a clear front-facing photo, or capture one with your camera. You get a top match, a match score for all ten shapes, and the exact ratios that produced them. Go to the calculator.
Accuracy note
The result is an estimate. Shape names are a styling convention, the thresholds here are heuristics rather than statistically validated cut-offs, and photo results depend on angle, lighting, hair over the hairline and facial hair. The manual and photo methods measure slightly different points, so they can disagree on close calls.
Calculate your face shape
Pick a method. All three feed the same six-ratio comparison, so the result looks the same whichever you choose.
Examples are only a guide to the format, not a target. Every face is different, and only the ratios matter.
How to measure at home
- Pull your hair back and face a mirror straight on, chin level.
- Use a flexible tape or a ruler; measure to the nearest millimetre if you can.
- Face length: from the centre of the hairline down to the bottom of the chin.
- Forehead: widest point between the temples. Cheekbones: widest point below the eyes. Jaw: corner to corner. Chin: across, a little above its tip.
- Repeat each measurement twice and use the average. Only the ratios matter, so the unit does not.
The camera only starts when you press Open camera, and it is released as soon as you capture or stop. Your browser will ask for permission first.
Your result
Your recent results
Only the shape name, score and date are saved, in this browser’s local storage. Photos are never stored.
How the face shape analysis works
The same classifier runs for every method. It does not guess from the picture; it compares numbers you can read.
Get five widths and a length
Typed in by you, or measured between landmark points in your photo.
Turn them into six ratios
Length ÷ cheekbone, forehead ÷ cheekbone, jaw ÷ cheekbone, chin ÷ jaw, forehead ÷ jaw, plus jaw angle.
Score ten reference profiles
Each ratio is compared with a profile for oval, round, square, rectangle, oblong, diamond, heart, triangle, pear and inverted triangle. Closer ratios earn more of the weight.
Show the gap
You see every score, plus the lead over the runner-up. A narrow lead is reported as a close call, not as a firm answer.
Understanding the 478 facial landmarks
The photo and camera methods use MediaPipe Face Mesh running in your browser through TensorFlow.js. Google documents this family of models as outputting an estimate of 478 three-dimensional face landmarks[1]; in the TensorFlow.js package, those 478 points are 468 face-mesh points plus 10 iris points, and the extra iris points are returned when iris refinement is switched on[2]. This page turns that option on and checks the length of the returned list: the result panel reports the real count it received, and only labels it “478 landmarks verified” when the count is exactly 478.
More landmarks do not make the shape label more accurate by themselves. Accuracy depends on the ratio thresholds, which are heuristics.
Only six of those points are used for the measurements. The iris points are not used, because they say nothing about the face outline. Read more about how facial landmarks work.
Which facial measurements matter?
Hand measurements follow the usual craniofacial landmarks: trichion (hairline), zygion (widest cheekbone point), gonion (jaw corner) and gnathion (chin tip)[3]. In photo mode the closest mesh points stand in for them, which is why the two methods can differ slightly.
| Measurement | By hand | Photo mode (mesh point) |
|---|---|---|
| A · Face length | Hairline to chin tip | Forehead top (10) to chin bottom (152). Point 10 sits a little below the true hairline, so length is slightly under-read. |
| B · Forehead width | Temple to temple | Upper face contour, left and right (54, 284) |
| C · Cheekbone width | Widest cheekbone point | Face contour at cheek level (234, 454) |
| D · Jaw width | Jaw corner to corner | Jaw contour (172, 397) |
| E · Chin width | Across the chin | Lower contour beside the chin (148, 377) |
| Jaw angle | Chosen from the jawline menu | Angle at the jaw point between the cheek point and chin point, averaged over both sides |
AI detector, quiz or measurement calculator?
People use these names interchangeably, but the methods differ. These are general distinctions; this page implements the first and third.
| Criterion | AI detector | Face shape quiz | Measurement calculator |
|---|---|---|---|
| Primary input | A photograph | Your answers to questions | Facial measurements |
| Main method | Landmark detection, then ratios | Rule-based questions | Geometric ratios |
| Main advantage | Measures automatically | Quick, no photo or tape | Every number is explicit |
| Main limitation | Sensitive to angle, lighting and model error | Depends on how you read your own face | Depends on how carefully you measure |
| Best for | A fast first estimate | Guided self-assessment | Checking proportions yourself |
| On this site | This page (Photo tab) and the face shape detector | Face shape quiz | This page (Measurements tab) |
If methods disagree, look at the ratios rather than the label. A face near the boundary between oval and oblong can land on either side depending on a 2 mm difference in forehead-top placement. Repeating the test with a straighter photo, or measuring by hand twice, usually settles it.
How accurate is a face shape calculator?
Treat the result as styling guidance, not a diagnosis. Three things limit accuracy here, and the page does not hide them.
- The categories are conventions. Oval, square, heart and the rest come from hairstyling and eyewear practice, not a formal medical taxonomy, and sources draw the boundaries differently.
- The thresholds are heuristics. The reference profiles and weights in this tool have not been calibrated against a labelled dataset. The “match score” is a weighted closeness measure from 0 to 100, not a probability, and the ten scores are independent so they do not add up to 100.
- Photos add error. A turned head, tilted chin, wide-angle selfie, hair over the forehead, glasses glare or a beard all move landmarks. The page warns you when it detects a turned head or a face cut off by the frame.
Landmark documentation tells you what the detector outputs. It is not evidence that the classification is right, and this page does not claim an accuracy percentage.
The ten shapes this calculator scores
Short descriptions with styling directions. For fuller guides, see types of face shapes, hairstyles by face shape, glasses by face shape, beard styles by face shape and celebrity face shape examples.
Who is it for? Anyone choosing a haircut, glasses, a beard shape or makeup placement who wants a proportion-based starting point. The calculator does not ask for gender and the ratios work the same for everyone; only the styling choices differ. People also search for a “face shape calculator for girls” or for men: the measurements and method are identical.
Frequently asked questions
Short answers about methods, privacy and accuracy.
What is a face shape calculator?
A face shape calculator estimates which named face shape (oval, round, square, heart, diamond and so on) your face is closest to by comparing ratios such as face length to cheekbone width, forehead to cheekbone width and jaw to cheekbone width.
How does the photo analysis work?
A face landmark model finds points on your face in the browser. The calculator measures distances between a handful of them, turns those into ratios, and scores the ratios against ten reference profiles. Photo analysis uses MediaPipe Face Mesh through TensorFlow.js with iris refinement enabled, which returns 478 landmarks; the page checks that count at run time.
How can I calculate my face shape with measurements?
Use a soft tape measure or a ruler against a mirror with your hair pulled back. Measure face length, forehead width, cheekbone width, jaw width and chin width, pick the unit, choose your jawline type, and press Calculate. Only the ratios matter, so any unit works as long as you use the same one for every measurement.
Is this face shape calculator free?
Yes. It is free, needs no account, and has no usage limit.
How accurate is a face shape calculator?
It is an estimate, not a diagnosis. Face shape names are a styling convention, and the thresholds used here are heuristics that have not been statistically validated. Results depend on photo angle, lighting, hair covering the hairline and how carefully measurements are taken. When two shapes score closely, the page tells you so instead of presenting a single answer as certain.
What are 478 facial landmarks?
MediaPipe’s face mesh model outputs 3D points across the face. With iris refinement on, that is 468 points on the face plus 10 on the irises, for 478 in total. The landmark count describes what the model detects; it does not by itself prove the face shape result is accurate.
Is my photo uploaded anywhere?
This page does not upload your photo. Detection runs in your browser. To do that, the page downloads the TensorFlow.js code and the face model from public hosts such as jsDelivr and TensorFlow Hub, so those hosts see an ordinary download request, but they do not receive your image. The page also loads its two fonts from Google Fonts on every visit.
Why might manual measurements and photo analysis give different results?
The photo method uses the landmark model’s forehead-top point, which sits slightly below the true hairline, and its cheekbone and jaw points are approximations of the anatomical landmarks you would find by hand. Camera angle and lens distortion also change the ratios. Small differences can move a face between neighbouring shapes such as oval and oblong.
Can I use the calculator without a camera?
Yes. You can upload an existing JPG, PNG or WEBP photo, or skip photos completely and enter measurements by hand.
Sources and further reading
Each source is listed with the specific claim it supports. Face shape names and the ratio profiles used for scoring are this site’s heuristics, not findings from these sources.
- Google AI Edge. Face landmark detection guide (MediaPipe Face Landmarker). Supports: the face mesh model outputs an estimate of 478 3D landmarks, and the task runs on single images or video.
- TensorFlow.js models. face-landmarks-detection package (README, v1.0.6 as bundled by jsDelivr). Supports: MediaPipeFaceMesh returns 478 keypoints per face with x, y and z; the code’s default is
refineLandmarks: false, which is why this page sets it to true and verifies the count itself. - Farkas, L.G. Anthropometry of the Head and Face, 2nd ed., Raven Press, 1994. Supports: the names of the standard craniofacial landmarks (trichion, zygion, gonion, gnathion) used in the manual measurement guide. Supports nothing about this tool’s face shape thresholds.
- MDN Web Docs. MediaDevices: getUserMedia(). Supports: the camera needs a secure context (HTTPS), always asks the user for permission, and fails with NotAllowedError, NotFoundError or NotReadableError in the cases the page reports.
Related face shape guides
Keep exploring once you have your result.