yogalike

How it works

From camera to one kind suggestion

No magic and no black box. Here is every step between your camera and the sentence on the screen — and the few things yogalike deliberately does not do.

1. Your camera, on your device

A pose model (MoveNet, from Google, the same one our sister apps use) runs inside this browser tab. It finds seventeen points on a body; yogalike immediately drops the five on the face and keeps twelve — shoulders, elbows, wrists, hips, knees, ankles. The video itself is never uploaded, recorded or stored.

2. Normalizing: your body, not anybody's

Raw positions would be meaningless — you might be two metres from the camera or four, tall or small, filming in portrait or landscape. So the points are re-centred on your hips and measured in units of your own torso length. After that, a stance is "1.6 of your leg lengths wide", not "300 pixels".

3. Measuring what a teacher looks at

Each pose defines a handful of measurements. Six kinds cover everything so far:

Poses are data, not code: adding a new asana means describing its measurements and reference ranges, and the same engine reads it. Sided poses like Tree and Warrior II are read both ways round, and yogalike uses whichever side you are actually doing.

4. Reference ranges, not one "correct" shape

Every measurement has a reference range, wide enough for real bodies. Anywhere inside it scores 100. Outside, the score eases down — 40 points per step of tolerance — so being a little outside costs a little. Each pose also comes in variations (a shorter stance, a foot on the calf, bent knees in Downward Dog), each with its own ranges: different ways of doing the pose well.

5. A score you can take apart

Measurements are grouped into legs, hips, torso, arms, and balance & stability. Your alignment is the weighted average of the measurements the camera could see — an unseen knee is left out, never marked down. Stability comes from how still your body is over the last two seconds, and it counts toward balance. The results show every measurement with its range, so you can check any number by hand.

6. One suggestion at a time

The suggestion on screen is the one that would add the most to your alignment right now. It stays put for a couple of seconds so you can act on it, and when you fix it, it says so. While you are holding well, it gets out of the way.

7. Holding, not counting

Yoga is not repetitions, so yogalike follows a state instead:

The overall score for a practice is 80% your average alignment while holding and 20% how much of the target hold you completed.

What an account keeps

If you are signed in, each practice saves its scores, every measurement against its range, the suggestions made, and the twelve body points five times a second — hip-centred, no face — so you can replay it as a stick figure. Delete any practice from your progress page and it is gone.

What an invitation carries

After a practice you can invite a friend to try the same pose. The picture is drawn in this tab from the pose's reference figure — not from your camera — with the pose's name and how long you held it; your average alignment only if you choose to add it. The link is the pose's ordinary page. Neither holds a body point, a measurement, a session or anything about your account, and nothing is saved to make them.

What yogalike is not

It is practice guidance, not medical advice. It cannot see your spine's curve, your feet, your breath or how a pose feels, and it cannot tell whether a pose suits your body. Its suggestions are adjustments toward a reference range — never a diagnosis, and never a promise about injury. Move within what feels comfortable, and if something hurts, stop and ask a qualified teacher or health professional.