Camera is off
Press Start Camera. Stand back far enough that your full body fits in frame.
Best for turnout and crossover checks: face the camera squarely, feet where the camera can see both.
Model not loaded yet.
on target
adjust
correct this
can't see clearly
--
/ 100 rolling
Kick Log — this session
Kicks will appear here at peak height with a form snapshot.
0
Sessions
--
Avg Score
0
Kicks Logged
Practice history
Style
Expressive arms
Off = traditional competitive standard (arms straight, still, at the sides). On = relaxes the arm-stillness check for styles/choreography that use arm movement.
Recalibrate stance
Clears your saved body-scale baseline. You'll be prompted to calibrate again next session.
Install this app
Checking install status…
Your data
Practice history
Stored only in this browser, on this device. Nothing is uploaded — export or clear it any time.
How the algorithm works
This is a practice mirror, not a certified adjudicator. It uses Google's MediaPipe Pose Landmarker running entirely on your device — nothing is uploaded — to estimate 33 body landmarks per frame, then turns joint geometry into the cues real teachers watch for:
- Turnout — angle of each foot's heel→toe line relative to your hip line. True turnout originates at the hip, so treat this as a proxy, not a hip X-ray.
- Crossover — how far the front ankle passes the midline/opposite hip in 3D space, mirroring the "diamond between the ankles" check.
- Kick height — vertical ankle position relative to your calibrated standing hip height, normalized to your own leg length.
- Toe point vs. "kicking your seat" — separates a bent-knee heel-to-glute pattern from a genuinely extended, pointed kick.
- Straight legs — hip–knee–ankle angle; flags the "sitting" bent-knee pattern.
- Arms — frame-to-frame wrist displacement (stillness) plus elbow angle (straightness).
- Shoulders / posture — shoulder-line level and shoulder-hip lean, most reliable from a slight side angle.
- Full Performance mode — runs every metric continuously and grays out any reading whose landmarks aren't clearly visible that frame, instead of assuming one fixed camera angle. It's built for routines where you travel and turn, but a metric can only be as good as what the camera can actually see at that instant.
Monocular video can't perfectly recover 3D hip rotation or depth. Use this alongside your teacher's eye, not instead of it.