Visual-motion synthesizer, live

Hybrid: word rules, a word model and your assistant

Saved Nov 19, 2026, 5:28am

Type a look, or ask your assistant for one in your own words, like make it feel like geometry or calmer. Words it knows apply at once; a word it does not know goes to a small word model that finds the closest word it does know. Keep this open while you talk: the badge says live when your assistant can change it.

Demo mode: moving on its own, no audio yet.

Audio

Words it knows

Mix words in one line. Words it doesn't know are skipped and named under the box.

Rules for mixingWhat happens when words disagree.
Shapes: 11 patternsSix are new in this version.
Colors and palettes21 colors and 15 palettes.
Speed, strength, amount, lookSet one slider or switch each.
Moods, plain EnglishEach sets several things at once.
SlangMapped to what each word means, many from AAVE.
Changing what's therePlain-English nudges and switches.

How it works

How audio drives itYour browser measures loudness in pitch ranges; nothing is uploaded.
What it is and isn'tHand-written rules, not a trained model.