Every gesture maps to a specific sound parameter. The left hand sculpts the atmosphere — a drone pad, a vocal ostinato, the reverb amount. The right hand plays the melody.
The audio engine is fully synthesized in the browser through the Web Audio API. No backend roundtrip, no latency beyond the camera frame.
Google on-device hand landmarker. 21 keypoints per hand, stable inference, GPU-accelerated. Confidence thresholds and a temporal consistency filter reject false positives.
A custom graph of oscillators, filters, convolution reverb, and feedback delay — all running natively in the browser, sample-accurate and zero-latency.
An evolving pad inspired by the "Forsitan Modulare" - Draen module, a monophonic sample-based laser harp, and a looped vocal ostinato.
The left hand controls the atmosphere. The right hand plays the melody. Each gesture below is wired to a specific parameter of the audio engine — move your hand, hear the change instantly.
Hold your hands in the frame, in front of the camera. Smooth movements produce smooth sound.

Fig. 1 — Six gestures, two hands. The left hand sculpts the drone; the right hand plays the melody.
Open PDFYou will be asked for camera access. The audio engine starts on your input only — we require a user gesture before any sound can play.