Face Shape Detector Troubleshooting

This is the most common message, and it means the model could not place its landmarks, not that anything is broken. The usual causes are a face that is turned or tilted rather than square to the camera, hair covering the forehead or jaw, a head cropped out of frame, or lighting too dim or too harsh for the features to read.

The fixes match the causes. Face the camera straight on with the lens at eye level, pull every strand of hair off your forehead, ears and jaw, and make sure your whole head fits in the frame from hairline to chin. Light your face evenly from the front, near a window if you can. A blank or non-face image will also return this message, which is expected. Get the face square, uncovered and well lit, and the landmarks place cleanly. The how it works page explains why these points matter.

If you chose the camera option and nothing appears, the cause is almost always a permission. Browsers block camera access until you allow it, so a prompt should appear the first time; if you dismissed it or blocked it earlier, the lens stays dark.

To fix it, reload the page and accept the camera prompt when it appears. If no prompt shows, open the site permissions in your browser, find the camera setting for this site, set it to allow, and reload. On a phone, also check that no other app is currently using the camera, since only one app can hold the lens at a time. If the camera still will not open, use the upload option instead and pick a photo from your device, which runs the identical analysis. Camera access is only ever used to capture a frame locally; nothing is recorded or sent.

A message about the model failing to load points to the one-time download. The detector fetches a runtime and a model file of roughly 4 megabytes on your first use, and a weak or interrupted connection can stall that fetch. After it downloads once, the browser caches it and later visits skip the step.

The fix is to check your connection and reload. Switch from mobile data to Wi-Fi if you can, or move somewhere with a stronger signal, then try again. An aggressive ad blocker or privacy extension can occasionally block the download, so if the problem persists, try disabling extensions for the site or opening it in a normal browser window. The download is a standard file fetch, and once it completes the model runs locally with no further network calls.

If the detector names a shape that does not match what you see, the photo is almost certainly the reason, since the measurement itself is reliable. A raised or lowered chin changes apparent face length and can turn an oval into a false round or oblong. Hair over the jaw hides the mandible corner and can soften a square into a round. A turned head skews the width readings.

Retake the photo with the camera level, the hair fully back and the expression neutral, then run it again. Take two photos and compare: if they now agree, the first one broke a rule. Remember too that a low confidence score is not an error but a sign your face sits between two shapes, in which case reading both neighbor pages is the right move. The accuracy page explains what moves a reading and what the score reports.

Slowness is nearly always the first load, not the analysis. The model download of about 4 megabytes takes a few seconds on a fast connection and a little longer on mobile data, and the scan animation covers the wait. Once the file is cached, later analyses start right away, so the tool should feel quick after the first use.

If it stays slow after that, the device or the browser is usually the limit. Close other tabs to free memory, especially on a phone, and make sure your browser is reasonably up to date. An older device runs the model a little slower but still finishes in seconds. If a single analysis seems to hang rather than just being slow, reload the page, which clears any stuck state without losing anything, since nothing was saved.

A few problems trace to the browser itself. The detector needs a modern browser that supports WebAssembly and, for the camera, media access; current versions of Safari, Chrome, Edge, Firefox and Samsung Internet all qualify. An outdated browser may not run the model, and the fix is to update it or switch to a current one.

Private or incognito windows can also interfere, since some block the storage the browser uses to cache the model or read a chosen file, which can cause repeated slow loads. Using a normal window resolves that. If a corporate or school network filters the model download, try a different network. These cases are uncommon, and a mainstream, up-to-date browser on an ordinary connection runs the tool without any of them.

Face shape detector troubleshooting is mostly a matter of the photo, a permission or the connection. A "no face found" message means the capture hid the landmarks, a dark camera means a blocked permission, and a stalled model means the one-time download needs a better connection. A wrong-looking result is a photo problem the same fixes solve, slowness is the first load and clears once cached, and the rare browser snag is solved by updating or leaving private mode. Clear the hair, allow the camera, reload on Wi-Fi, and run a level photo through the detector; the photo guide shows exactly what a clean capture looks like.