The technical foundation

The hand check uses MediaPipe Hand Landmarker, a freely available model by Google that locates up to 21 points on a hand: wrist, finger joints and fingertips. From this point cloud we compute lengths, angles and ratios. The model and the computing library are served from our own server and cached by your browser - your image is processed locally and never uploaded.

What is actually measured

Eight values form the basis of the report. Each is geometry - what the tradition reads into them is a separate question.

  1. Palm width to palm length - wide and angular versus narrow and slender.
  2. Finger length to palm length - long, airy hands versus compact ones.
  3. The 2D:4D ratio - index finger length divided by ring finger length, measured from the base joint to the tip. Below 0.97 the ring finger is longer, above 1.00 the index finger; in between counts as balanced. The thresholds are deliberately rough.
  4. The thumb angle - flexibility at the thumb joint.
  5. Finger spread - how wide the fingers stand apart.
  6. The heart-line curve - estimated from the curvature of the palm.
  7. The life-line arch - how far the life line sweeps around the thumb pad.
  8. Handedness - reported by the model, shown in the result.

How the four hand types emerge

The four-type system comes from four-element philosophy, not from natural science. The tool maps it using two values: palm shape (square vs. rectangular) and finger length (short vs. long).

  • square palm, short fingers → Earth
  • square palm, long fingers → Air
  • rectangular palm, short fingers → Fire
  • rectangular palm, long fingers → Water

Between the axes there are no hard jumps: a hand between square and rectangular, or between short and long fingers, is weighted proportionally and shown as percentages - an approximation, not a measurement.

The hand check sees no palm lines

The model finds joints, not creases. That is why the result works even if your hand sits in shadow - but it also means that every "line" statement in the report is a geometric approximation based on the 21 points, not a pattern read from your photo.

What else is not robust

  • Rough thresholds: the cut-offs for hand types and the 2D:4D bands are orientation, not calibration.
  • Two dimensions: a tilted hand shortens fingers and distorts ratios. The tool warns when it detects a tilt.
  • One sample: a single photo is not an average. Repeats under the same light give more stable values.
  • Skin and age: creases change with age, work and moisture - the same hand can look different over time.
  • Handedness: the model can misclassify left and right, especially with unusual angles.

What actually has a research base

  • 2D:4D: the best-studied hand measure, linked to prenatal hormone exposure - with small effects and mixed replication.
  • Crease formation: major hand creases form before birth and change with age and mechanical load.
  • Hand shape: partly heritable; proportions vary between populations.
  • Not supported: character, talent, health or future predictions from hand lines.

How to check reliability yourself

  1. Take three photos under identical light and compare the eight values.
  2. Analyse both hands and compare - differences are normal.
  3. Read the descriptions without looking at the labels first: does the text fit you, or does it fit almost anyone?
  4. Change one variable at a time (light, angle, distance) - not all at once.

Our editorial rules

  • Separate measurement and interpretation. When we talk about proportions, that is verifiable. When we assign a trait to a hand type, that is traditional symbolism - and we say so.
  • No future, no diagnosis. No statements about lifespan, illness, fertility or events.
  • Sources instead of gut feeling. Where research exists, we link it; where it does not, we say so openly.
  • No fear-mongering. The life line myth is the best example - and the reason we phrase these pages so carefully.

Questions about the method

Why does the model need about 17 MB?

The hand model and the WebAssembly computing library are downloaded once (about 17 MB) from our server and then cached by your browser. On the first analysis step it loads; after that, analyses start immediately.

Why are the line readings approximations?

The model locates joints, not creases. The line values in the report are computed geometrically from the 21 hand points - they describe the shape of the palm, not the pattern of your lines.

Why is the hand type shown in percentages?

Because real hands sit between the two axes. Instead of forcing a hand into one of four boxes, the tool weights the four archetypes proportionally. The top value is your main type; the rest shows the mix.

Is the result reproducible?

Mostly - if the photo is comparable. Take three photos under the same light and compare the values; large jumps usually mean the hand was tilted, cut off or partly hidden.

Privacy

The tool works client-side. The model is downloaded once from our server and cached. No image, no landmarks and no analysis data are transmitted to us or to third parties. Details in our privacy policy.