SkillMirror

How a practice session actually works

You watch a short phrase, record yourself dancing it, and the phone itself measures your movement against the reference — then tells you one honest thing to work on. These are the real screens, as built today.

What happens the first time you open it?

A one-minute welcome teaches everything before the camera exists: what the product does, why it needs the camera, where to stand, and what feedback will sound like. It is skippable, and you can replay it any time.

The first-run welcome: a chartreuse movement trace above the words “A mirror with a measure”.
The first screen a new dancer sees in SkillMirror, as built.

How do you set up a recording?

Corner brackets show the frame the analysis needs — whole body, head to feet, light on you, facing the camera. A three-second countdown gives you time to step back. If the analysis cannot see part of you, it says so afterwards; it never guesses.

The welcome pane teaching framing: a line-figure dancer standing inside four corner brackets.
The framing lesson — the same corner brackets later frame the live camera.

What happens while you record?

The phone records your attempt against the phrase’s own length. Nothing is analysed live and nothing is uploaded — every frame is measured on the phone after you stop.

Recording in progress: a large clock reading 0:05 of the 4:00 phrase over a progress bar.
The recording screen. Analysis runs on the phone after you stop — the screen says exactly that.

What do the results look like?

One fix leads, in a dancer’s words — which joint, which direction, which moment of the phrase. Every measurement behind it stays one tap away: the exact degrees, the sections, and what the analysis could not see. When your own earlier attempt exists and a change clears the measurement’s noise floor, the improvement is acknowledged — measured, never invented.

A results screen leading with one coached fix and a measured-improvement panel, with a “Show the measurements” button.
Results, as built: real engine output computed from a synthetic practice capture.

What is the one fix?

The single clearest signal in your attempt — the deviation that reached furthest beyond that joint’s own measured reporting floor. Other findings are set aside on purpose: one thing at a time.

The Focus Fix screen: one coached instruction above the exact measurement it came from.
The one fix, with the measurement behind it — real engine output on a synthetic capture.

Where does practice happen?

Anywhere a phone can stand: a studio at midnight, a plaza, a corner of home.

A dancer mid-move in a worn Seoul practice room at night, a propped phone running the app. AI-generated scene · real app screen
A late practice room in Seoul. AI-generated scene · real app screen
A street group at dusk gathered around a propped phone running the app. AI-generated scene · real app screen
A session at dusk. AI-generated scene · real app screen
A boxer shadow-training beside a propped phone running the app. AI-generated scene · real app screen
Shadow rounds, counted honestly. AI-generated scene · real app screen
A person mid-squat in a small home training corner, a propped phone running the app. AI-generated scene · real app screen
A training corner at home. AI-generated scene · real app screen
A performer mid-song in a Paris apartment, a propped phone running the app. AI-generated scene · real app screen
A song rehearsed in Paris. AI-generated scene · real app screen
A street dancer on a Rio plaza, a propped phone running the app. AI-generated scene · real app screen
A plaza session in Rio. AI-generated scene · real app screen

Illustrative scenes, AI-generated with the real interface shown; today the app teaches dance phrases, and the wider creation engine is the roadmap’s next chapter.

Why only one fix, and why in degrees?

Every joint has a measured floor below which a difference cannot be told apart from you simply repeating yourself. The app only speaks when a deviation clears that floor, ranks what it says by how far past the floor it reached, and leads with the strongest signal. There is no overall score, because no honest one exists for a body learning a phrase.

Screens on this page are from the app as built today. Results screens show real engine output computed from a synthetic practice capture — the voice and layout are exactly what a dancer sees; the numbers are not a real person’s.

The product is in development and is not yet available in the app stores.