How to take a good selfie for face analysis — and why the photo decides your number

Why the photo decides the number, not your face

The photo decides the number because every face tool measures the picture, not you — it has no way of knowing which distances belong to your skull and which ones the camera invented. The same face shot at 25 cm and at a metre gives two different sets of proportions, and both look equally real to a landmark model.

Distance is the biggest of them. Up close, the nose sits proportionally much nearer the lens than the ears do, so it prints larger while the jaw and the sides of the face print narrower; step back and that gap collapses. Nothing about the face changed — only the geometry the camera sampled.

Light is the second. A lamp directly above your head drops shadow into the eye sockets and under the cheekbones, which is exactly where a puffiness or a cheek reading looks; a window behind you turns the whole face into a silhouette with no readable edges at all.

Tilt moves everything at once: chin up shortens the midface, chin down lengthens it, and a head turned a few degrees off the lens shortens one half of the face against the other — which reads as asymmetry that is not in the face.

Expression is the quickest accident of the four. A smile lifts the cheeks, narrows the eye opening and moves both eye corners, so the symmetry and the eye-corner angle you measured belong to the smile rather than to you.

The checklist for a face-analysis selfie

The checklist is ten items, and every one of them is free. Work down it once, note where you stood, and the rest of this page is only the reasoning behind it.

• Camera at eye level. Phone upright, lens level with your eyes — not raised above you, not resting on your chest.
• Arm’s length, fully extended. Better still: prop the phone at about a metre and use the timer, which takes most of the close-up distortion out of the frame.
• Face a window. Daylight in front of you, roughly a step back from the glass, so the light lands on the whole face at once.
• Kill the overhead lamp. Ceiling light above your head puts shadow exactly where the eye and cheek readings look.
• Neutral face: lips together, no smile, no jaw clench, eyes open and looking at the lens.
• Glasses off. Frames sit across the eye corners the tools measure, and a reflection can hide an eye completely.
• Hair off the forehead, and back off the temples and jaw — a fringe or a hood hides the outline the shape and symmetry tools trace.
• One person in the frame. Two comparable faces and the tool stops rather than guess which one is yours.
• No filter, no beauty mode, no retouching — and no crop that clips the top of the head or the chin.
• Same room, same window, same hour, next week. This is the item that turns a photo into a measurement.

The best lighting for a face photo

The best lighting for a face photo is soft daylight coming from in front of you: face a window, stand about a step back from it, and switch off anything directly overhead. Front light fills the eye sockets and the area under the cheekbone instead of carving shadow into them, which is what makes two photos of the same face comparable.

Three setups ruin a reading, and all three are common. A ceiling lamp above you writes dark hollows under the brow and the cheekbone. A window or a bright screen behind you leaves a silhouette the landmark model has to guess at. And mixed light — warm bulb on one side, cold daylight on the other — lights the two halves of your face differently, which is exactly the difference a symmetry tool is looking for.

The LOOKSMAXBRO app says the same thing the moment a frame goes dark. Its capture screen prints “Not enough light”, and under it: “turn to a window · lamp behind the phone”. The teaching screen before it puts the rule on a tile — LIGHT · window in front.

How far the camera should be, and what it does up close

The camera should be at arm’s length at minimum, and about a metre if you can prop the phone and use the timer. A front camera does distort your face at close range — not because the lens is bad, but because perspective is: at 25 to 30 cm the nose is proportionally far nearer the lens than the ears, so it enlarges and the face narrows around it.

The app measures this rather than trusting you. Its capture screen prints a live DISTANCE readout and asks for 28 to 70 cm, saying “Too close · arm out a little” or “Too far · bring the phone in” until you are inside that window. Arm’s length sits comfortably in the upper half of it.

The browser tools have no distance sensor, so they check the consequence instead: a face narrower than about a fifth of the frame is refused as too small to measure honestly. There is no upper limit — which is precisely why holding the phone too close is your problem to fix rather than the tool’s.

This is also the honest answer to why you look worse in selfies than in the mirror. Most of that gap is 25 cm of perspective plus a face you have only ever seen flipped; a metre of distance returns proportions much closer to the ones other people see.

What the free browser tools need from a photo

The face tools on this site need one face, turned to the lens, big enough in the frame, and a file the browser can open — and each of them says which of those failed rather than quietly measuring something else.

• One face. Two faces of comparable size stop the run so the tool never picks for you; a much smaller face in the background is measured around, with a warning printed beside the number.
• Head straight to the lens. A slight turn shows the number with a caveat; past that, no number is shown at all, because projection rather than your face would be most of the answer.
• The face at least about a fifth of the frame wide, or the tool answers that it is too small to measure honestly.
• A file the browser can decode: JPEG, PNG, HEIC and WebP all work.
• Eyes open, no glasses, camera at eye level — the line printed on the drop zone of every face tool.

What the app scan adds to the same rules

The LOOKSMAXBRO app scan adds two things a browser tool cannot: readouts that check the frame while you shoot, and a second shot at 45°. The rules themselves are the ones above — the app just refuses to let you guess whether you followed them.

Its teaching screen states the whole standard in four tiles: LIGHT · window in front, CHIN · level, HOUR · morning, FACE · neutral. Under them, one line — “The four readouts on the capture screen check exactly these — you never guess.”

In the viewfinder those readouts speak in short corrections: “Level the chin · phone upright · chin down a touch”, “Face in the oval”, “Too close · arm out a little”. When the frame is finally right the line becomes “Hold it.”, with “shoots on 3 · hands off” underneath it, and the shutter fires by itself; once the count is running the screen switches to “Hold still.” — so pressing a button cannot nudge the phone at the moment of capture.

Then the second frame: “Phone out at 45°, eyes straight ahead — that is the shot.” The side view is where the jaw and cheek read from an angle no front photo contains. The HOUR · morning tile matters for the same reason a weekly comparison does — a face is at its puffiest in the first hour after waking, so the app asks for that hour every time rather than an average of whenever you remembered.

Where the photo goes: nowhere

The photo is processed on your own device and is never uploaded — in the browser tools and in the app alike. On this site the landmark model is downloaded into your tab, runs there on the image you picked, and hands back numbers; the page’s own connection policy limits every network call it can make to this site’s own origin, so the photo cannot be sent to any other server, and nothing in the tools sends it to this one either.

The model is MediaPipe’s Face Landmarker: 478 points on the face, served from this site’s own origin rather than a third-party network, loaded the first time you actually pick a photo. Nothing is stored either — reload the tab and the image, the landmarks and the number are all gone.

The app works the same way with more of it: face processing and scoring run on the iPhone, and its own gallery screen states the rule plainly — “Scoring still runs on this iPhone, and photos never leave it.” The only way a photo travels is if you deliberately share it yourself.

Same conditions every week, or you are measuring the light

Shoot the weekly photo in the same spot, at the same hour, in the same light — otherwise the difference between two scans is the room and not your face. This is the single rule that separates a photo you can compare from a photo you can only look at.

The app states it as its whole capture doctrine: “Same light. Same angle.” — and, underneath, “The score compares your photos with each other — so the frame must be the same every day.” Nothing in the checklist above matters more than repeating it identically.

It is also why the score re-scans weekly rather than daily, and why the first number arrives only after scans on seven separate days: sleep, salt, the light in the room and the angle you held the phone at move a face reading further in one night than honest work moves it in weeks. A weekly delta per metric survives that noise. A daily number does not.

What a good photo cannot fix

A good photo cannot make the number mean more than it does: a low score from a perfect frame is still an aesthetic, entertainment-style estimate — not a diagnosis, not a medical assessment, and not a fact about you. It is a model’s guess at how a panel of human raters would have ranked one picture.

What the photo can do is remove every excuse. When distance, light, tilt and expression are fixed and repeated, a change between two scans is a change in you rather than in the weather — which is the only condition under which a number is worth acting on at all.

And when it is worth acting on, the useful question is never the score. It is which of the things that actually move a face you are willing to do for a few weeks, and which ones are marketing.