The short answers, without the hedging.

No. The image is decoded in your browser, analyzed by a local model, and kept in memory for the session. The app has no image upload endpoint and does not call an external AI API.

Measurements, landmarks, shape scores, the visible skin-tone estimate, and the date. The photo is not included. The saved report stays in localStorage on that device until you delete it or clear site data.

No analytics or advertising scripts. The app uses browser storage for the theme preference and, if you choose, the saved report. It does not use tracking cookies.

Yes. Close the tab to clear the in-memory photo and session report. Use “Delete saved report” on the results page to remove the localStorage report. Your browser’s site-data settings can clear everything, including the theme preference.

It is a geometric estimate from a 2D photo, not a verified label, and no trained face-shape image classifier is used. The report shows the ratios, the closest alternatives, and a capture-quality score. Head pose, lens distortion, hair, glasses, and lighting can shift the result. There is no published clinical benchmark behind these specific thresholds, and no per-class or per-demographic accuracy is claimed.

The measurement and prototype logic is gender-neutral and does not classify gender. The men’s and women’s guides are editorial presentation layers, and their reference portraits are AI-generated; they are not validation data. There is no representative, independently labeled validation set, so the app does not claim separate male or female accuracy. The data audit and the requirements for a future trained model are documented on the methodology page.

The percentage is a relative geometry fit, with the closest reading indexed to 100. It is not a probability that you “are” that shape. Real faces often sit between categories, and the second reading can be close.

Use the manual correction on the report. It changes the primary label without changing the measurements. You can also try another photo with neutral light and a straight head pose.

No. It does not score beauty, age, gender, ethnicity, health, or identity. It measures visible proportions for styling decisions.

A front-facing photo with the whole head in frame, even light, open eyes, a neutral expression, and no heavy filters. Remove glasses if they sit away from the face or obscure the cheek line. Avoid strong side shadows and extreme camera angles.

The local MediaPipe runtime uses SIMD instructions for acceptable performance. The app checks support before loading the model. Current Chrome, Edge, Safari, and Firefox generally support it; older browsers get a clear message instead of a broken analysis.

Once the detector page is open and the local model has loaded, analysing another image continues to work without a network connection. A full page reload or a fresh tab still needs the static files from the site; this build does not install an offline app shell, so offline reloads are not guaranteed.

Yes. The layout is responsive, and the file input offers a camera capture option where the browser supports it. The first load may be heavier because of the model and WASM runtime.

The app samples cheek and forehead pixels, filters unreliable values, takes the median colour, converts it to CIE Lab, and calculates ITA. It reports depth, undertone, and a confidence note. Lighting, makeup, filters, and camera processing can change the result.

They are deterministic rules based on your shape, visible undertone, and optional quiz answers. They are not generated by an AI model and they are not a dress code. Personal preference and comfort come first.

Yes. The measurement ratios and the specific styling direction are more useful than the label alone. Share the report or the JSON export, not the photo unless you choose to.