Scroll to dive in

Sound made visible

NSKU makes audio-reactive 3D visuals: geometry that listens. Every shape below is driven, right now, by the sound you hear. Move your cursor through it. Click to play a note.

Frequency Icosphere(01)

A subdivided sphere where every vertex is tied to a frequency bin. The bass swells the core; the hats shiver across the skin.

  • 7 bands
  • FFT 2048
  • Baked lighting

Waveform Topology(02)

Time becomes terrain. Each slice of the waveform scrolls back into a landscape you fly over.

  • Rolling history
  • Height = energy

Reactive Fractal(03)

A system that never repeats. Every kick pushes the parameters, so it never lands in the same place twice.

  • GPU shader
  • 1.4 ms / frame

Depth Point Cloud(04)

Kinect depth, cleaned and re-lit as particles. A face that breathes with the vocal stem, not the whole mix.

  • Kinect
  • Depth cleanup
  • Vocal stem

Polyhedra Array(05)

Platonic solids in a harmonic ring, each tuned to its own band and pulsing on its own subdivision of the beat.

  • Beat grid
  • Phase-locked

Fiber Nebula(06)

Spiral arms of light, spun by the bass and lit by the hats. A galaxy at club tempo.

  • Stem-split
  • Bloom + trails

How it works

Inside the engine

Most music visualizers react to volume: louder means bigger. The NSKU engine listens the way a mixing engineer does. It pulls a track apart into instruments, measures each one many times a second, and knows where the next beat will fall before it arrives. The result is motion that feels played, not triggered.

  1. 01

    Capture

    Audio files (MP3, WAV, FLAC, OGG), live microphone, or the soundtrack of a video. Video frames stay locked to the audio clock within one frame.

  2. 02

    Stem separation

    Demucs v4 splits the song into vocals, drums, bass and other, cached per file. Each part of the image follows its own instrument: the face follows the voice, the ring follows the kit.

  3. 03

    Spectral analysis

    30 times a second: a 2048-sample Hann-window FFT, 7 bands from sub-bass to presence, RMS energy and spectral centroid. Adaptive floors and ceilings keep quiet and loud tracks using the full range.

  4. 04

    Beat grid

    Spectral-flux onset detection and tempo estimation build a grid of where beats will land, so a punch impulse fires on every kick, even between detected beats.

  5. 05

    A/V sync

    The visuals follow what leaves the speakers, not what was just queued. A phase-locked clock tracks when each buffer reaches the DAC, so a hit lands on the frame you hear it.

  6. 06

    Render

    Velocity-driven physics, GPU shaders and geometry uploaded once. 60 fps on macOS, 30 fps on a Raspberry Pi 5 for installations.

  • 0Stems separated
  • 0Frequency bands
  • 0Analyses / second
  • 0Frames / second

Questions

What are audio-reactive visuals?

Graphics generated or animated in real time from a sound signal. The audio is analysed many times per second (loudness, frequency bands, beats) and those measurements drive shape, motion, color and camera.

How is this different from a regular music visualizer?

Most visualizers react to overall volume. NSKU separates a track into stems first, so each element follows its own instrument, and it schedules motion on a phase-locked beat grid so hits land on the frame you hear them.

Can I use my own music?

Yes. Drop any audio file on this page or use your microphone and every visual follows it. Nothing is uploaded: the analysis runs in your browser.

What does it run on?

This gallery runs in any modern browser with WebGL. The NSKU visualizer app runs on macOS at 60 fps and on Raspberry Pi 5 at 30 fps for installations and live shows.

Your turn

Play it loud

Swap the soundtrack for your microphone or any track in your library. Everything here will follow it.

Music: Pharoelidae, “Vector Equilibriums” · Head scan: “Infinite” by Lee Perry-Smith, CC BY 3.0 · © NSKU