About FaceReport

FaceReport is an independent, self-funded collection of fourteen free face analysis tools — scoring pages such as the PSL scale and PSL score, eye geometry pages like the canthal tilt and hunter eyes checkers, proportion workbenches for the golden ratio, beauty ratios and facial thirds, a symmetry triptych, a face shape detector and two live online mirrors. This page explains who we are, how the tools actually work, and the editorial and privacy rules every page follows.

Who builds FaceReport

The site is built and maintained by FaceReport Research, a small independent team with backgrounds in front-end engineering, geometry and technical writing. We have no investors, no advertising partners and no data business — which is easy to verify: the site has no accounts, no newsletter, no tracking pixels and no photo upload endpoint at all. The project exists because face-measurement tools are popular while honest, private, well-engineered ones are rare, and we thought the gap was worth closing.

Our editorial stance is deliberately plain: we measure, we explain the arithmetic, and we refuse to dress the results up as science they are not. Where internet communities attach harsh labels to measurements — "negative canthal tilt", low PSL scale positions, "prey eyes" — our pages deliberately use neutral, geometric language and say plainly that a photo measurement is not a verdict on a person.

How the tools work

Every analysis tool on the site follows the same pipeline, and the whole pipeline runs inside your browser:

1. Photo input — you upload a JPEG, PNG or WebP file, or capture a frame with your camera. The file is decoded in memory; it is never transmitted anywhere.
2. Landmark detection — a self-hosted face landmark model (Google's open-source MediaPipe Face Landmarker, about 4 MB, cached after first use) locates 478 points on the face, entirely on your device.
3. Geometry — transparent TypeScript functions turn those points into angles, ratios and symmetry offsets. The formulas, weights and display tolerances are published on each page and versioned, so the same photo always yields the same result under the same version.
4. Rendering — the annotated overlay and result cards are drawn in the page. Downloads are generated locally, only when you click.

The two mirror tools use your camera through the browser's standard permission prompt and never record anything: the stream is displayed live and released the moment you stop or leave the page.

What we will not do

Some things are cheaper to promise than to build, so we built the site so the promises are verifiable:

• No uploads. There is no server-side photo endpoint in the code base.
• No invented percentiles. Pages never claim you beat "87% of users" — no such comparison exists or could honestly exist here.
• No medical or beauty verdicts. Outputs are labeled as heuristic photo geometry for entertainment and learning.
• No dark patterns. No login walls, no credit prompts, no watermarks, no "share to see your result".

Editorial standards

Each tool page carries a visible byline and update date, a direct-answer summary under its heading, a transparent methodology section, an honest limits section, and a sources list citing the references we relied on — from Wikipedia articles on the golden ratio and facial symmetry to Google's MediaPipe documentation and MDN's camera API reference. Where a claim on our pages depends on an external source, the source is linked; we would rather cite than assert.

Scores follow a published, versioned configuration: symmetry, spacing and thirds components are mapped to 0–100 with public formulas, averaged with fixed weights (0.4 / 0.3 / 0.3), and mapped onto the illustrative 1–8 PSL-style scale as 1 + 7 × index ÷ 100. When a component cannot be measured honestly — an unconfirmed hairline, a rotated head — the tool says so instead of guessing a filler value.

Languages

FaceReport is published in six languages: English (the reference language), French, German, Spanish, Russian and Indonesian. Each page is written, not machine-translated, with its own locally researched primary keyword — the German PSL Skala page is a German page, not a translated English one. Facts, formulas and privacy guarantees are identical across all languages.

Frequently asked questions

Is FaceReport really free? Yes — every tool, every language, no account and no premium tier. The site costs little to run precisely because it computes nothing on servers: pages are static, and the analysis runs on your device.

Can I trust a browser-based measurement? For what it claims, yes. The landmark model is a widely used open-source component, the geometry is plain arithmetic you can audit in the page source, and the quality gates refuse to output precise numbers when the photo cannot support them. What no tool can do is measure a three-dimensional face from a two-dimensional photo — and our pages say so explicitly.

Why do results change between photos? Because lighting, lens, camera height, distance and expression all move real measurements. Run the same tool on two or three photos and compare: the spread you see is the honest error bar of photo-based geometry, and understanding it is more useful than any single number.

Who owns my downloaded results? You do. Files you download are generated in your browser and saved by your click; they contain nothing about other users (there are no other users' data at all) and nothing about you beyond the pixels of the card.

Contact

We read everything sent our way. To keep this site free of accounts and trackers, contact runs through the repository and the site itself rather than a newsletter funnel: open an issue or a pull request in the public project repository, or mention @facereport where you found us. For privacy questions, read the privacy policy first — it is short, and most answers are "nothing is collected, so there is nothing to delete".

If you are reporting a measurement that looks wrong, include the photo conditions (camera height, lighting, expression) — those factors move real measurements, and reproducing them is the fastest path to a fix.