Camera
Fix Guide
Face Shape Detector Not Working? Here’s How to Diagnose It
The face shape detector usually works on the first try — but camera permission errors, unexpected results, and inconsistent readings across sessions are common enough that a dedicated troubleshooting reference is useful. This guide covers everything that can go wrong and exactly how to fix it.
Issues fall into two categories: the camera won’t load at all (access problems), or the camera loads but the result doesn’t look right (quality problems). The sections below cover both — plus a re-testing methodology for cases where you’re not sure if the result is accurate.
Why the Same Person Gets Different Results
If you’ve used the detector more than once and got different face shape results, you’re not alone. The AI measures geometric ratios from pixel data — which means any change in lighting, camera angle, hair position, or head position changes what it measures. Your face shape hasn’t changed; the input data has.
Here are the five variables that cause inconsistent results most often, and the specific measurement each one affects:
Camera above eye level compresses the forehead and makes the face look wider relative to its length — often causing oval to read as round. Camera below eye level does the opposite. Even 10–15cm above eye level is enough to shift the classification.
Shadows on one side of the face affect where the jaw and cheekbone landmarks are detected. Different lighting between sessions — morning window vs. evening lamp — changes which zones are shadowed and shifts those specific measurements.
If your hair is pulled back more thoroughly in one session than another, the forehead width and jaw edge measurements will differ. Even a few strands over the hairline affect the forehead reading.
A slight tilt introduces asymmetry into every width measurement — the model reads a different jaw width on the raised side versus the lowered side. A 5-degree tilt is enough to shift the classification in borderline cases.
A smile raises the cheeks and shifts the jaw contour. If you smiled in one session and had a neutral expression in another, the soft-tissue measurements will differ — sometimes enough to affect the shape category.
How to interpret inconsistent results
How to Use the Camera Feature
Follow these steps in order. The preparation steps (1–3) have more impact on your result than the capture steps (4–6) — get the environment right before you sit down.
Prepare your environment before opening the app
Choose your position before launching the camera. Sit or stand facing a window or an evenly lit wall — light should come from in front of you, not from behind or from one side. If using artificial light, use two sources at roughly equal angles on either side to avoid single-source shadows. This setup takes thirty seconds and has the largest impact on result accuracy of any step.
Pull hair back and remove glasses
Use a clip or tie to pull all hair away from the forehead, temples, and jaw — these are the zones where facial landmarks are measured. Glasses frames create edges that interfere with landmark detection around the brow and eye area. Both should be done before opening the app, not after the preview loads.
Open the detector and grant camera permission
Navigate to the face shape detector and select the camera option. Your browser will request camera permission — click Allow. If you've previously denied permission, see the troubleshooting section below. The live preview will load once permission is granted.
Position your face using the alignment guide
Hold your device at eye level — not tilting up toward you or away. Align your face with the oval guide so your hairline sits near the top edge and your chin near the bottom edge. Your face should fill the guide without the guide cutting into the hairline or jaw. Look directly at the camera, not at your own image on screen.
Check the live preview for even lighting
Before capturing, look at the preview image. Both sides of your face should appear equally lit with no dark zones at the temples or jaw. The image should not look washed out or overexposed. If you see uneven lighting, move toward or away from the light source until both sides balance.
Capture with a neutral expression and review
Hold still, look directly at the camera lens, relax your face into a neutral expression (lips closed, jaw relaxed, no smile), and press the capture button. Your result appears within seconds. If the shape classification seems unexpected, review the result against the troubleshooting criteria in Section 04 before retaking.
Lighting and Positioning: What Each Variable Affects
Most camera setup guides tell you what to do without explaining why. Understanding the specific impact of each variable helps you make the right call in non-ideal conditions — for example, knowing whether to proceed with imperfect lighting or whether it's worth waiting for better conditions.
Conditions That Improve Accuracy
- ✓Even frontal lighting — balances both sides of the face, allowing the detector to accurately locate jaw and temple landmarks on both sides equally
- ✓Natural window light (overcast) — diffuse cloud-filtered light is the most even natural light source — direct sun through a window creates strong directional shadows
- ✓Camera at exact eye level — eliminates the perspective distortion that makes the forehead appear proportionally larger (camera below) or the jaw appear wider (camera above)
- ✓Face fills the alignment guide — more pixels per facial landmark = more precise measurements; a face that is too small in frame loses measurement precision
- ✓Neutral expression with relaxed jaw — a smiling or tensed expression shifts the cheek and jaw muscles, altering the measured soft-tissue contours
- ✓Hair fully cleared from hairline and jaw — the detector measures from actual skin edges — hair covering these zones forces estimation rather than measurement
Conditions That Degrade Accuracy
- ✕Single directional light source — creates a shadow on one side of the face, obscuring the jaw contour and temple landmarks on that side
- ✕Backlight or window behind you — the camera exposes for the bright background, leaving the face underexposed and landmarks difficult to locate
- ✕Camera tilted upward or downward — even a 10-degree tilt creates significant perspective distortion in the forehead-to-jaw ratio — the single most important measurement
- ✕Head turned or tilted to one side — a rotated head means one side of the face is measured at a different angle than the other, producing asymmetric landmark placement
- ✕Glasses frames on during capture — frame edges are detected as facial contours by the landmark model, particularly around the brow and cheekbone zones
- ✕Face too small in frame — sub-pixel landmark measurements introduce rounding errors that compound across the four key facial measurements
“The camera doesn’t lie — but its lighting does. Most ‘wrong’ results are accurate readings of a poorly set-up photo.”
The Three Lighting Setups, Ranked
Diffuse natural light from a window in front of you
Sit facing a window on an overcast day. The cloud cover acts as a natural diffuser, creating even, shadowless light across the full face. This is the easiest high-quality setup and requires no equipment. On a sunny day, avoid direct sun rays hitting one side of the face — step back from the window so you're in the reflected light rather than direct sunlight.
Two artificial light sources at 45 degrees on each side
Two desk lamps or ring lights placed at roughly equal angles on either side of you — not directly in front — create fill light on both sides of the face. If you only have one lamp, bounce it off a white wall or ceiling to diffuse it. The goal is eliminating the shadow on one side, not adding dramatic lighting.
Single overhead room light with a white reflector
A single overhead light creates a downward shadow under the jaw and nose. If this is your only option, hold a white sheet of paper or a book in your lap to bounce light back upward under the chin and jaw — this softens the shadow on the jaw contour zone, which is one of the most critical measurement areas.
Troubleshooting Camera Issues
Camera issues fall into two categories: access problems (the camera doesn't load at all) and quality problems (the camera loads but the result is unexpected). Each has a distinct diagnostic path.
Access Problems
Permission denied — camera blocked
The browser has remembered a previous denial. In Chrome: click the padlock icon in the address bar → Site settings → Camera → Allow. In Firefox: click the padlock → Connection secure → More information → Permissions → Use the Camera → Allow. In Safari: Settings → Websites → Camera → find the site → Allow. After changing the permission, reload the page.
No camera detected — permission dialog never appeared
This usually means another application has exclusive access to the camera. Close video call apps (Zoom, Teams, Meet), streaming software (OBS), and any other browser tabs with camera access. On Windows, check Task Manager for processes with camera activity. On Mac, check the camera indicator light — if it's on with no visible application, force quit from Activity Monitor. Restart the browser after closing other apps.
Camera loads but shows a black screen
Common on laptops with privacy shutters — check if the physical camera cover is slid open. On some devices, a driver issue prevents the browser from receiving the video stream; restarting the browser or the device typically resolves it. If the issue persists across browsers, the camera driver may need updating via Device Manager (Windows) or System Report (Mac).
Camera permission granted but preview shows wrong camera
Devices with both front and rear cameras may default to the rear camera. Look for a "flip camera" or "switch camera" button in the interface — on mobile, this toggles between cameras. For face shape detection, always use the front-facing camera. On desktop, if you have multiple camera devices, the browser's camera selection can be changed in the browser settings.
Camera works on another site but not here
The site may be blocked specifically. Check browser settings for site-level camera permissions and ensure this site is set to Allow rather than Ask or Block. Also check if a browser extension (ad blocker, privacy extension) is blocking the camera API — disable extensions temporarily to test.
Quality Problems — Unexpected Results
If the camera worked but your face shape result seems wrong, use this diagnostic sequence before retaking:
Was there a shadow on one side of your face?
If yes, one side's jaw or temple landmarks were likely estimated rather than measured. Move to better lighting and retake.
Was your head tilted, even slightly?
Tilt is the most common cause of unexpected oval-to-round or round-to-square misclassification. Look at the captured frame — if your head is even slightly off-vertical, retake with extra attention to keeping your head straight.
Did your hair cover any part of the forehead, temples, or jawline?
Hair over the hairline or jaw causes the model to underestimate the width of those zones. Re-clip the hair more thoroughly and retake.
Were you wearing glasses?
Glass frames interfere with brow and eye-area landmarks. Remove them and retake.
Were you smiling or had a tense expression?
Muscle movement shifts soft tissue. Relax the jaw fully and retake with a neutral expression.
Was the face too small in the alignment guide?
Move closer to the camera so your face fills most of the guide area, and retake.
If the Result Is Consistent Across Multiple Good Photos
How to Do a Reliable Re-Test
If your first result was unexpected and you want to get a definitive answer, follow this controlled re-test protocol — it eliminates the variables that cause inconsistency and gives you a result you can trust.
Choose a consistent location
Pick a spot you can return to: ideally a chair facing a window where you control the lighting. Using the same spot for all retests eliminates lighting variation as a variable. Note the time of day — morning and afternoon window light differ significantly as the sun angle changes.
Standardize your preparation
Hair in the same position for every test — ideally pulled fully back into a bun or clip. No glasses. The same neutral expression. These are the variables that most commonly change between casual retests without the user noticing.
Set the camera to exactly eye level
Prop the device on a fixed surface at eye level — a phone stand, a stack of books, or propped against a monitor. Do not hold the phone yourself, because arm angle varies between captures. The camera-at-eye-level requirement is the single most important technical condition.
Take 3 captures, not just 1
Take three photos in quick succession without moving, then choose the one where the face looks most centered and evenly lit. Don't take one photo, submit it, then retake if you don't like the result — that introduces selection bias toward the result you prefer. Choose based on photo quality, not result.
Run two full sessions on different days
Your face changes slightly with hydration, sleep, and time of day. Two sessions on separate days, both under the same controlled conditions, give you a comparison that accounts for natural daily variation. If both sessions agree, the result is reliable.
“If two carefully done sessions agree, trust the result — even if it doesn’t match your expectation. The AI measures geometry, not self-perception.”
Borderline Results: When You’re Between Two Shapes
If controlled retests consistently alternate between two face shapes — say, oval and oblong, or square and rectangle — you are genuinely borderline between those categories. This is a normal, valid result. Most face shapes exist on a spectrum rather than falling cleanly into one category.
The practical solution: read the styling advice for both categories and apply the recommendations that overlap between them. The overlap is your most reliable guidance — it works for your face regardless of which side of the boundary you’re on.
Frequently Asked Questions
Does the camera feature work on all browsers?
Is my video stream recorded or stored?
My result was oval last time but round this time. Which is right?
Can I use a photo from my camera roll instead of taking a new one?
What resolution does the camera need to be for good results?
Does the camera use flash?
Further Reading
Naeem Ullah
Founder, Face Shape Detector • AI & Facial Proportion Researcher
Founder of faceshapedetector.app · 4+ years in facial proportion research · 200,000+ monthly readers
Naeem Ullah is the founder of Face Shape Detector and has spent over four years researching how facial landmark geometry translates into practical styling decisions. His work draws on training principles from professional hairstyling, optician certification programs, and academic literature on facial symmetry and proportion. He built the face detection system at the core of this tool and personally writes and reviews every styling guide published on this site. His guides are read by over 200,000 users monthly across 140+ countries.