Free Face Shape Detector: Can You Use It Without Uploading a Photo?

Yes. In Face Shape Lens, choose Enter measurements and enter your forehead width, cheekbone width, jaw width and face length. You can get a face-shape estimate without selecting a photo. If you prefer the photo option, the current tool processes the selected image inside your browser. The website still loads files and uses visit analytics.

Want to keep your photo out of the process?

Measure straight across each width, rather than following the curve of your face. Use the same unit for every entry. Face length runs from the center of the actual hairline to the bottom of the chin, not from the top of your hairstyle.

The result compares those proportions with the tool's shape rules. It is a styling estimate. The classifier has not been independently validated, and the displayed rule distances are not accuracy percentages.

We tried the no-photo route before asking for your measurements

These checks used invented numbers, not anyone's face measurements, in the live Face Shape Lens tool on September 22, 2026. We used the Codex in-app browser, left the photo picker empty and did not enable a camera. The first calculation used the site's manual-calculation browser tool; subsequent changes used the visible form. This checks the available workflow, not classification accuracy.

A result without a portrait

With forehead 14 cm, cheekbones 15 cm, jaw 12 cm and length 22 cm, Face Shape Lens returned Diamond, with Heart and Oval listed as close alternatives. Its displayed length-to-cheekbone ratio was 1.467; the other displayed ratios were 0.800 for jaw-to-cheekbone and 0.933 for forehead-to-cheekbone. The three rule distances were Diamond 0.0000, Heart 0.0222 and Oval 0.0278. No image selection was needed to reach that result.

What happens while you change an entry?

Changing only length from 22 to 24 cleared the displayed result before we pressed the calculation button. After pressing it, the tool returned Oval and showed Diamond, Heart and Oblong as close alternatives. It displayed a length-to-cheekbone ratio of 1.600; jaw-to-cheekbone and forehead-to-cheekbone stayed at 0.800 and 0.933. The nearest distances became Oval 0.0000, Diamond 0.0167 and Heart 0.0222. The interface still marked the result as a close call.

Hiding a result did not clear our entries

We switched from Enter measurements to Use a photo after the second calculation. The result disappeared and the photo picker was still empty. Switching back to Enter measurements showed the same four values: 14, 15, 12 and 24. The result area remained empty until another calculation. In this test, changing tabs hid the classification but kept the typed measurements in the form.

We then replaced length with 100, leaving the other entries at 14, 15 and 12, and pressed the calculation button. The form showed an instruction to check the units and measurement points; it did not display a face-shape result. The value 100 remained in its input. Rejecting a calculation therefore did not erase the value that caused the error.

Finally, we reloaded the page and reopened Enter measurements. All four input boxes were blank and no classification was displayed. That is the reload behaviour we observed in this browser session. We did not test browser crash recovery, autofill, back-button restoration or synchronised browser history.

What the site's current code does with this route

The inspected manual-form handler reads the four number fields and passes them to the local classifier. It has no photo-file requirement. The classifier rejects inputs when the largest entered value divided by the smallest exceeds four; our rejected example produced 100 ÷ 12 = 8.333…, above that limit. This is an input-consistency rule in this implementation, not a medical boundary for faces.

The site's current analytics configuration builds the recorded page location from the origin and pathname, leaving out the query string and fragment. It reduces the supplied referrer to its origin and disables Google signals and advertising-personalisation signals in that configuration. The manual form does not add its measurements or classification to an analytics event in the code inspected. This is a code finding, not a packet-capture audit of every browser request.

Sources for the code checks: manual form and reset handlers, classifier and analytics configuration, checked September 22, 2026. The live checks above are separate from the six providers' published policy statements below.

What appeared in the network record?

The initial request list contained the homepage document, two site stylesheets, the application script, the classifier module and the site's analytics script. A separate request targeted Google's tag-manager script. A Cloudflare Insights beacon script also appeared, but the browser recorded that particular request as blocked by the site's content security policy. Its presence in the request list therefore did not establish that the beacon had loaded or sent data.

A Google Analytics request carried the event name page_view. A later request carried scroll and a percent_scrolled parameter of 90. Both targeted the Google Analytics collection endpoint using POST. The browser recorded an aborted loading event for each request; this trace does not confirm delivery to Google. We distinguish the attempted requests from successfully received analytics records.

The recorded analytics request parameters included the page address, page title, language, screen resolution, browser information and client and session identifiers. The scroll request also included a scroll percentage and engagement timing. Neither of those two recorded parameter lists included a forehead, cheekbone, jaw or face-length field, or a Diamond classification. The captured request objects contained no recorded POST body. We have omitted identifier values from this article.

For the interval starting immediately before we opened the manual tab and ending after the result appeared, the browser event reader returned no new request events. The Diamond result was nevertheless visible with all four submitted numbers. The later scroll request appeared outside that interval. Our observation is thus narrower than saying that pressing the button prevents subsequent analytics activity.

The recorded request list through that later scroll event did not contain the site's photo-analysis worker or landmark-model files. The application code starts that worker inside the photo-loading function; the manual handler calls the classifier directly. That explains why the observed no-photo calculation did not require downloading those photo-analysis resources. It does not establish that a first visit works offline: the page and its manual-classifier scripts still had to load.

Could the already-loaded manual form recalculate offline?

After the online calculation, we enabled offline network emulation for that browser tab. With forehead, cheekbone and jaw entries still at 14, 15 and 12 cm, we changed length to 24 cm and pressed the calculation button. The result updated to Oval, with Diamond, Heart and Oblong as close alternatives, while the offline setting was active. The displayed length ratio updated to 1.600. We then restored the tab's normal online setting.

The offline test started with the page already loaded. It did not test loading the website for the first time without a connection, and it did not test the photo route offline.

What do six other detectors say about photos?

Download the six-tool comparison (HTML)

Official policy and tool pages opened in the live browser. No portraits submitted, cameras activated, registrations completed, or server deletion tested. “Not found” means not found on the inspected pages, not that a practice or feature is absent. Retention of photos, measurements and paid-report data is kept separate.

Dates use America/Los_Angeles. Each table cell contains the collection date and source links. Quote IDs open the exact excerpts below. A stated event is not converted into a guessed number of hours.

Read the promise for the feature you intend to use

The linked VisuTry policy addresses virtual try-on; its detector page describes local image handling. Reversely’s policy identifies reverse search in its image-storage sentence. Facewow’s Face Data clause identifies AI Avatar features. Those distinctions prevent a retention rule for one service from silently becoming a claim about another.

The Face Report distinguishes the free detector’s photo from its saved measurements and classification. Its result-retention period cannot be presented as the lifetime of an uploaded detector photo. Face Shape Lab’s homepage storage statement and its policy’s performance-improvement wording should both remain visible; neither supplies a detector-photo deletion deadline in the pages inspected.

These are comparisons of published statements. We did not test other providers’ network traffic or deletion systems, and an unexplained field is not evidence of misconduct.

Exact source excerpts

Short excerpts retain the original wording; some are sentence fragments or button labels. Each ID links the table to its source. These are provider statements, not verified server behaviour.

What happens when you choose a photo?

You then check the points used for measurement. The hairline needs your attention because the initial point is an upper-forehead estimate. Move it to the actual center of your hairline before accepting the measurements.

Does local processing mean there are no network requests?

No. Your browser requests the page, scripts and model files. The site also has analytics for visits and active users. The current analytics setup does not attach your chosen photo, entered measurements or face-shape result to its events.

What does Remove photo clear?

The button clears the tool's selected-file field and displayed image, closes its active image bitmap, stops the photo worker and resets the measurement points and result. It does not delete the original file from your device.

It also does not erase visit statistics already sent while you browsed. Clearing the photo workspace and clearing analytics records are different operations.

What should you check before choosing another detector?

Start with its explanation of photo handling. Does it say where analysis happens, whether images are retained, and whether they have another use? If the page only says “private” or “secure,” those questions are still unanswered.

A file picker alone cannot answer them either. It lets the page access a file you select; the website's code determines what happens next. Do not assume every detector behaves like this one.

If you are not comfortable selecting a face photo, leave it unselected. On Face Shape Lens, the measurement tab gives you a way to explore your proportions without adding an image to the tool.