Shape a sound
Explore 36 presets and five synthesis engines. Drift, Sub, Drive, Cutoff, Resonance, Release, Echo and Space give you direct control of the timbre.
Obsidian.
I build synthesizers to explore what each new AI model can make. Obsidian, developed with ChatGPT and Codex, turns that recurring experiment into an instrument: play a sound, build a groove and connect a MIDI controller.
Build a scene
Four editable chord pads, voicing options and an arpeggiator meet a four-part, 16-step drum sequencer. Record one, two or four bars of keyboard or chord-pad notes into a performance loop.
Invite the AI
The Neural Atelier can suggest a whole scene, sculpt a patch or evolve your current settings. It changes instrument parameters; Web Audio generates the sound in the browser.
Bring your hands
Play with your mouse, touch, computer keyboard or USB MIDI. The MPK mini Plus profile maps four pads to drums, four to chords and eight knobs to sound controls. MIDI Learn adapts the mapping.
Keep the idea
Save and load scenes as JSON, return to the last scene in your browser or record the audio output. Your performance loop can travel with the saved scene.
Test by playing
Sound generation, musical timing, interaction and visual feedback have to work together. This is my recurring creative test for new models, with a useful instrument as the result.
Using the tool
Next in the LabCanvasStart at a low listening level and enable audio in the instrument. For USB MIDI, use Chrome or Edge, connect your controller and allow MIDI access. AI features need an internet connection and have usage limits; ordinary playing does not need an AI request.