Streams
Interactive granular synthesis with AI text-to-audio, six visualization modes, and up to 256 concurrent grains.
Six interaction modes
Each mode renders audio as a different particle system.
Drag to push · Double-click for a vortexFlow
Spectral droplets meander down a flowing stream
Flock
Curl-noise murmuration drifts and splits
Flame
Rising tongues of spectral fire with additive glow
Fractal
Golden-ratio spiral expanding from center
Flora
Spectral blossoms bloom on growing branches
Flood
Raindrops fall and splash into spectral arcs
Create new material for resynthesis with AI
Describe a sound and generate source material locally via CoreML, then navigate the latent space between your generations to decode new hybrids on demand.
Text-to-Audio
Describe what you want to hear and the model generates ~12 seconds of stereo audio at 44.1kHz.
Fast and local
Runs on Apple Silicon via CoreML. No cloud, no internet. 8-step rectified flow sampling generates audio in seconds.
Latent space explorer
Every generation drops an anchor in the model's latent space. Click between anchors to decode hybrid textures that blend what you've already made.
Navigate, don't just generate
- Every prompt becomes an anchor — a point in a 2D projection of the full latent space.
- Click between anchors to decode a new ~12s sample at that position, blending the nearest results.
- Recompute the projection as your collection grows and new clusters emerge.
- Everything stays local on Apple Silicon via CoreML — no uploads, no waiting on the network.
Granular sampling
Composite grains combine multiple bandpass-filtered components into a single voice. Triggered, continuous, or sequenced.
Triggered
Click the stream to sample nearby particles into composite grains. Drag velocity controls timbre richness.
Continuous
Auto-spawn grains at a configurable rate with position spread across the stream.
MIDI & Sequenced
Trigger grains from MIDI notes or the built-in step sequencer. Scale-aware routing tunes grains to musical intervals.
Studio-quality processing
Resonant strings, granular delay, algorithmic reverb, and character shaping.
Algorithmic Reverb
Dattorro plate, FDN, nested allpass, and velvet noise tanks. 18 presets with controls for mix, decay, and size.
Granular Delay
Flange, chorus, and delay ranges with digital, analog, and tape character modes. Hermite-interpolated reads and feedback filtering.
Resonator Bank
Resonant strings with scale-aware tuning, adjustable energy and damping. Excited by audio input or the step sequencer.
Character Compressor
Drive, tone shaping, mid EQ, and output compression at the end of the signal chain.
Route anything to anything
Route LFOs to any parameter, including other LFOs. Two sequencer modes for rhythmic and gestural control.
Multi-shape, multi-target
Sine, triangle, square, saw, sample-and-hold. Assign to any parameter. BPM-sync or free-running.
Rhythmic grain triggering
Program melodic patterns on a grid. Scale-aware note routing drives grains and resonator strings. Syncs to host tempo.
Gestural recording
Record sequences of gestures and manipulations to the audio stream in real time. Each step captures pitch, velocity, and timing from your interaction with the visualization.
Real-time sound processing
Lock-free audio, parallel grain rendering, zero-copy ring buffers.