About

This site does one thing: it measures the proportions of a face from a photo and names the shape those proportions describe. It runs entirely in your browser, it is free, and it requires no account.

Why it exists

Face shape is genuinely useful for a narrow set of decisions: which hairstyle adds width where you want it, which glasses frames balance which part of your face, whether a beard should be wide or long. Most tools that answer the question do so with an upload button and a confident percentage, and tell you nothing about how they got there.

So this one takes the opposite approach on two points. It never takes your photo — the analysis happens on your device, and there is no upload endpoint to breach. And it publishes itsmethod and its limitations, including where the result is likely to be wrong, so you can decide how much to trust it.

How it is built

• A face landmark model runs locally in your browser and returns 478 reference points on the face. It is downloaded once — about 3.6 MB, plus about 3 MB of runtime that executes it — and cached afterwards.
• Four measurements are taken from those points — face length, cheekbone width, forehead width and jaw width — and converted into ratios, so the result does not depend on how close you were to the camera.
• The ratios are compared against the ranges that define each shape. When two shapes score close together, both are shown rather than one being picked silently.
• The site is a static site with no server-side code. That is what makes the privacy claim structural rather than a promise: there is no server that could receive an image.

What it deliberately does not do

• It does not score attractiveness. Face shape describes bone structure. Judging how someone looks is a different product, and not one this site is interested in building.
• It does not collect anything. No account, no email, no analytics, no cookies, no stored result. See the privacy policy.
• It does not claim precision it cannot demonstrate. Themethod page says plainly what the tool cannot account for and what the published accuracy of this kind of classifier actually looks like.
• It does not use a contact form. A form needs a server, and this site has none by design. Contact is by email instead — see contact.

An honest note on accuracy

Face shape categories are conventions, not measured facts. Two people looking at the same borderline face often disagree about which label fits. Published work on automated face shape classification reports accuracy in the low-to-mid 80s on small labelled sets, and no tool in this category — including the well-funded commercial ones — publishes a validated accuracy figure for its consumer product.

So we measured ours and published the number. On 200 photos from a public dataset, the category we picked matched the dataset's label 30% of the time — 50% counting the runner-up. On a five-category problem, chance alone is 20%, so that is barely better than guessing, and the dataset's own labels are of unknown quality. The honest conclusion is that the four measurements are the reliable part of this tool and the single category label is not. That is why the measurements are shown first, why the runner-up is always displayed, and why themethod page says all of this rather than quoting a flattering figure. If you think it got your shape wrong, tell us — that is how the boundaries get fixed.

Contact

edrfhhokmjun@gmail.com · or see thecontact page for what is useful to include.