Using the Face Shape Detector on Mobile

The requirements are minimal: a modern mobile browser and a connection long enough to download the model once. Safari on iPhone and iPad, and Chrome, Edge, Firefox or Samsung Internet on Android all run the detector, since they all support the WebAssembly and camera features MediaPipe uses.

There is no app to download, which surprises people who expect a face tool to live in an app store. Everything happens on a web page, so you simply open the site and start. You do not sign in, hand over an email or grant any permission beyond camera access if you choose the camera option. An older phone works too, though a recent one runs the model a little faster. Once the model has downloaded on your first visit, the browser caches it, so later analyses start immediately.

You have two ways to feed the detector on a phone, and both keep the image on the device. The camera option opens your front-facing lens so you can take a photo on the spot, with the framing visible on screen as you line up. The upload option lets you pick an existing photo from your camera roll, which is handy if you already have a good front-facing shot.

The camera is usually the easier path on mobile, since holding the phone up at eye level gives you a clean, square capture and you can retake instantly if the framing is off. The upload path suits a photo someone else took of you, which avoids the slight reach of a selfie. Either way, the analysis is identical, and neither sends the picture anywhere. If you use the camera, allow the permission prompt when it appears, since the tool cannot open the lens without it.

Camera height and tilt are where phone captures go wrong, so this is the part to get right. Hold the phone so the lens sits level with your eyes, not below your chin looking up and not above your forehead looking down. A low angle stretches the apparent face length and a high one shortens it, and either skews the shape reading.

Keep the phone upright and square to your face rather than tipped back, and hold it far enough away that your whole head fits in frame with a little space around it. A too-close crop cuts off the hairline or the jaw, and the detector reads the outline, so the outline has to be fully in the shot. Propping the phone against something at eye level, or asking someone to hold it, steadies the frame and makes the level angle easier than a one-handed selfie. The photo guide covers the rest of the capture.

On a phone, the privacy story is the same as everywhere else on the site: nothing is uploaded. MediaPipe runs inside the mobile browser, so the photo and every measurement stay on the device, and there is no server that receives the image. When you close the tab, the photo is discarded.

That matters more on a phone, where the camera roll holds personal photos. Choosing a picture from your roll does not copy it anywhere; the browser reads it locally to place the landmarks and nothing more. Camera captures behave the same way, held in memory only for the length of the analysis. The absence of an upload endpoint is a deliberate, load-bearing part of how the site is built, and it holds identically on mobile and desktop. You are not trading privacy for the convenience of a phone.

The detector downloads a model file of roughly 4 megabytes the first time you use it, which is the one moment a connection matters. On mobile data that download takes a few seconds; on Wi-Fi it is near instant. After that, the browser caches the file, so every later analysis skips the download and starts right away.

Running the model on a phone takes a moment longer than on a fast laptop, but still finishes in seconds, and the scan animation covers the wait. A recent phone with a capable browser handles it comfortably. If the first load feels slow, it is the one-time model fetch rather than the analysis, and it will not repeat. The AI model page explains what the download contains and why it runs locally.

A few issues are specific to phones, and each has a simple fix. If the camera will not open, check that you allowed the camera permission and that no other app is holding the lens; reloading the page re-prompts. If the model seems stuck loading, a weak connection is usually the cause, so switch to Wi-Fi and reload.

If a face is not found, the capture is nearly always the reason: raise the phone to eye level, pull your hair back, and make sure your whole head is in frame with even light. If the page feels slow to respond, close other browser tabs to free memory on the phone. Most phone problems come down to permissions, connection or framing, and the troubleshooting guide covers the full list. Once the capture is clean, the phone reads your shape as accurately as any computer.

The face shape detector on mobile runs entirely in the phone's browser, with no app, no account and no upload, mapping the same 468 landmarks and four measurements it uses everywhere. You can shoot with the camera or pick a saved photo, both kept on the device; the angle to watch is camera height, held level with your eyes; and the only network moment is the one-time model download of about 4 megabytes. Privacy on mobile matches desktop exactly. Open the site on your phone, line up a level shot, and run it through the detector.