Face-ratio tools have two problems: most of them want the photo uploaded to a server, and many of them end with a score telling you how attractive you are. I wanted neither. So PhiRatio does one narrow job: you place the landmarks yourself, and the photo never leaves the tab.

I built PhiRatio around that constraint. You open a front-facing photo, place the points, and compare a few facial proportions to a golden-ratio and thirds-and-fifths grid. There is no attractiveness score, percentile, or overall verdict, and it is not a face-swap or face-edit tool. The tool is for people 18 or older.

The tool is free to use. There's a one-time $7.99 unlock for the watermark-free high-resolution PNG and the PDF report — no subscription. Every ratio, both measuring windows, the grids, the watermarked preview, and the SVG/PNG templates are free.

Automatic landmark detection is the obvious next step, but it usually means one of two things: the photo goes to a server, or the browser has to download a detection model. I didn't want either. Manual placement keeps the photo in the tab, avoids a model download, and makes every ratio explainable from the points you just put down. The cost is that placing those points is more work, especially on a phone — which is the part I most want feedback on.

The tool also has two vertical measuring windows, and they do not cover the same span, so they cannot share one reference value.

The hairline-to-chin window uses the golden ratio, φ = 1.618, as the face length / face width reference. A φ split of that window at the base of the nose also gives upper face / lower face a reference of 1.618.

The default window is shorter at the top: visible forehead to chin. Only the upper segment is shortened, by a named constant called the visible forehead offset, VFO = 0.238, measured in face-width units:

VFO = φ − 1.38 = 0.238 visible length / width reference = φ − VFO = 1.38 visible upper / lower reference = (1 − VFO) / (φ − 1) ≈ 1.233 

Those visible-window numbers are derived design references, not a standard of appearance. VFO itself is a shared design assumption, not a measured property of a given face. If you mix a 1.38 reading with a 1.618 reference, or the other way around, you are comparing two different constructions.

The measurement core is a pure function: landmark coordinates and a window id go in, and each ratio comes out as a number, or stays empty if a required point is missing. Two of the window-dependent rows look like this:

function valueFor(row, landmarks, frame, windowId) { const top = at(landmarks, frame, windowTop(windowId)); const menton = at(landmarks, frame, "menton"); const subnasale = at(landmarks, frame, "subnasale"); switch (row.id) { case "face-length-over-width": { const width = faceWidth(landmarks); // zygion to zygion const length = Math.abs(menton.y - top.y); return length / width; } case "upper-face-over-lower-face": { const upper = Math.abs(subnasale.y - top.y); const lower = Math.abs(menton.y - subnasale.y); return upper / lower; } // ...the other nine rows } } 

Vertical distances are taken after a roll correction that lines up the inner eye corners, so a slightly tilted photo does not quietly skew the up/down ratios.

If you try it, the tool is here: phiratio.app. The transparent PNG and SVG grids are on the templates page. If you've placed eight or more points on a phone, tell me which step felt most awkward — drag precision, pinch-zoom, or undo.