Vizzy turns a sentence into a moving picture: you type a prompt, a local LLM designs a visual patch, and the native engine plays it. This guide gets your first visual on screen and the sound driving it — then points at everything else.
Visuals are designed by a local LLM through Ollama — nothing leaves your machine. The first time you launch Vizzy, if it can't find a model it offers to install one for you: pick from the list and wait for the download. That's the whole setup.
Already have it sorted, or want to do it by hand? The setup steps on the home page cover installing Ollama and pulling a model manually. The LLM dot in the top bar is green when you're connected.
The screen is split into two live scenes, A (left) and B (right), each with four channels. Press CUE A so the four builder cards along the bottom are editing Scene A's channels.
In the first card, type something into the prompt box —
neon plasma tunnel pulsing with the bass — and hit
Generate. The LLM designs a patch (a generator
+ palette + motion + audio routing); a few seconds later it appears in
that channel's preview and on the main output.
Raise the channel's fader in the mixer to bring it into the picture (channel 1 starts up; the rest start down). Don't love it? Edit the prompt and Generate again, or tweak SCALE, W/H, TILT and the colour knobs on the card.
Tip: flip the little mode toggle on the card from GLSL to SCENE and Generate again to get a 3D fly-through (terrain or tunnel) instead of a flat shader. You can also right-click anything in the Library to drop it straight onto a channel.
Find the Audio group in the top bar, pick an input from the dropdown, then click the Live button — it turns green and the dot lights up.
To react to whatever's playing on your computer, choose Computer audio at the top of the list. On Windows and Linux that captures the system output directly; on macOS it routes through a virtual loopback device — install BlackHole (free) and the entry picks it up automatically (it'll say so if one isn't found). Prefer to react to the room instead? Pick your built-in mic or an audio interface.
Vizzy now measures four smoothed bands every frame — low, mid, high and overall level — plus a beat detector (separate kick / snare / hat layers) and automatic BPM, and feeds them to every visual. Most generated patches already react out of the box, so play some music and watch it move.
Want to steer how it reacts? Open a channel's AUDIO tab and set BAND (which band — or a beat layer — drives it) and AMT (how hard). Automation effects with the ♪ marker lit follow the same routing. The meter panel in the top bar shows every level and lets you tune the beat detector and sync the detected BPM.
No reaction? Check the device is actually receiving signal and the Live dot is green. macOS will ask for microphone permission the first time — allow it, then toggle Live off and on.
Press CUE B and build a second look on the right scene exactly the same way — it stays off the output until you bring it in, so you can prep B while A is live. The A–B crossfader at the bottom of the mixer blends the two scenes onto the master (double-click it to snap-flip).
Each channel also has an M mute and its own fader, so within a scene you're mixing four layers additively before the crossfade.
You don't start from a blank canvas. On first launch Vizzy seeds a
starter library so there's plenty to play with before you generate
anything of your own — open the Library panel to
browse it. Every example was made the same way yours will be, so each one
is also a working example you can pull apart. Look for the
Example · prefix.
It's all yours to remix. Every example is an ordinary Library entry, so tweak any channel, SAVE a new version, or delete the ones you don't want.
.glb / .gltf / .obj / .stl models,
.png / .jpg images and .mp4 / .mov / .webm
video clips straight onto the 3D / IMG / VID tabs..vizzy file. Reset App… wipes everything
back to a fresh, out-of-the-box install and reloads the bundled examples —
save first if you want to keep your work.
