How to Take a Selfie for a Skin Scan
Skinzy will not let you submit a photo it cannot read. This is what the camera is checking while you hold still, in the same numbers the code uses.
The Scanner Checks Before It Shoots, Not After
Most photo-based tools accept whatever you upload and hope for the best. Skinzy runs the quality check in your browser, live, before the shutter is available at all — so there is no way to submit a frame the server would turn down.
While the camera is open, a face landmarker runs on every frame and answers four questions: is your face large enough in the frame, is it turned too far, is it tilted too far up or down, and is anything covering it. If all four pass for eight consecutive frames, the photo is taken automatically.
That eight-frame requirement is why it feels like a short pause rather than an instant snap. A single good frame can be luck — a blink, a flicker of light, a frame caught mid-turn. Eight in a row is a held position. If a frame fails, the counter drops back by three rather than resetting to zero, so one wobble does not send you back to the start.
Framing: Fill More of the Frame Than You Think
Your face has to occupy at least 15% of the frame by area. That is the single most common reason a scan will not fire, and it catches people out because a photo that looks fine to you is often taken at arm's length from a wide front camera.
Fifteen percent by area is roughly the point where your face spans about two-fifths of the frame's width. In practice: bring the phone closer than feels natural, until your chin and hairline are near the top and bottom of the preview with a little room to spare.
The reason the bar is that high is resolution per zone. The scan divides your face into five regions and tiles each one into patches; a small face in a large frame leaves too few pixels in the under-eye and chin zones to measure anything reliably. The server will accept a face as small as 5% of the frame, but the browser deliberately holds you to three times that, because a photo that barely passes produces a report full of low-confidence readings.
Angle: Face the Camera, But Do Not Worry About Tilt
Turning your head left or right (yaw) is limited to 20 degrees. Tipping it up or down (pitch) is limited to 22 degrees. Both are strict, and for the same reason: a turned or tipped face foreshortens one side, so a cheek that is angled away reads as a different size and a different brightness than the one facing the lens. That asymmetry would land in your scores as a real difference between your left and right cheek, when it is only geometry.
Tilt — rolling your head sideways, ear toward shoulder — is treated completely differently. It is allowed up to 30 degrees, and it is not really a limit at all: the pipeline rotates the image and the landmark set together to straighten it before anything is measured. Roll is corrected, not rejected, because unlike yaw and pitch it does not change what the camera can see of your face. Only the extreme case is refused.
The practical version: look straight at the lens, keep your chin level, and stop worrying about whether your head is perfectly upright.
Obstruction: Hair, Hands, Glasses
A second model runs alongside the landmarker and segments skin from everything else. It is the only thing that can tell that your fringe is across your forehead or your hand is resting on your chin, and capture stays blocked until it has loaded — the shutter is disabled and the stable-frame counter is held below the threshold rather than allowed to fill.
Push hair back off your forehead and cheeks. Take off glasses. Nothing should sit between the camera and the skin being measured, because an obstructed zone does not produce a wrong score — it produces no score, and the report comes back with that region marked unmeasured.
Facial hair is handled separately and deliberately. A beard or heavy stubble makes the chin and jaw zone unreadable for acne specifically, so that axis is marked unmeasured for that zone rather than guessed at. This is the intended trade: a handful of dark follicles read as lesions to a classifier, and three of them were once enough to drive an entire routine.
Lighting: Even Beats Bright
There is no minimum brightness you need to hit, and a window is better than a ring light. What matters is that the light reaches both sides of your face roughly equally.
Side light is the problem case. A lamp or window to one side puts one cheek in shadow, and a shadowed cheek reads darker and lower-contrast than the lit one — which is exactly what a pigmentation or dryness measurement is sensitive to. Face the light source rather than standing beside it.
The scoring is more robust to overall lighting than people expect. Measured across eleven faces under exposure, gamma and colour-temperature shifts, the oiliness reading moves by about 0.05 points on the 0–10 scale, sensitivity by 0.11 and the overall composite by 0.21. What lighting does not fix is grain: in a dark room your phone raises ISO, and heavy sensor noise is the one thing that can make the acne reading collapse entirely. Bright, even, indirect daylight avoids both.
A Short Checklist
Face a window, or any soft light source, straight on. No lamp off to one side.
Hair back, glasses off, nothing touching your face.
Bare skin — no makeup, and ideally not straight after cleansing, which temporarily changes both oil and redness.
Bring the phone closer than feels natural, until your face fills most of the preview.
Look at the lens, chin level, and hold still for a moment. It fires on its own.
Questions People Ask
Almost always because your face is too small in the frame — it has to cover at least 15% of the frame by area. After that, the usual causes are turning your head more than 20 degrees to one side, tipping your chin more than 22 degrees up or down, or hair across your forehead. The on-screen prompt names whichever check is failing.
You need even lighting more than bright lighting. Face a window rather than standing beside one, so both cheeks get the same light. Across eleven test faces, changing exposure, gamma and colour temperature moved the overall score by about 0.21 points out of 10 — so ordinary lighting differences are tolerable. A dark room is the real problem, because the grain your phone adds at high ISO is what degrades the acne reading.
Skinzy captures through your camera rather than accepting an upload, so the quality gate can run before the photo exists. A mirror shot also puts glass, distance and usually side-lighting between you and the lens, all of which work against a measurement.
No. Makeup is designed to change exactly the surface properties the scan measures — it mattifies shine, covers redness and evens out pigmentation. Scan on bare skin, and wait a little after cleansing rather than photographing skin that has just been stripped.
Very little. Sideways tilt is corrected rather than rejected: the image and the facial landmarks are rotated together to straighten the face before measurement. Only a tilt beyond 30 degrees is refused. Turning left or right, and tipping up or down, are the angles that genuinely have to be kept small.