toy software · zero seriousness
VIZZY
A toy VJ instrument that's fun to play with: type a prompt, a local LLM dreams up the visual, a native GPU engine renders it, and you ride the faders while everything wobbles to the music.
- The lazy way: just open Vizzy. If it can't find an LLM it offers to download Ollama and a model for you, with a picker showing what each model needs (RAM, disk). Done. The steps below are only for doing it by hand.
-
Install Ollama — the local LLM
runtime Vizzy talks to. Nothing leaves your machine;
no API keys, no cloud.
# macOS download from ollama.com/download, or: brew install --cask ollama-app # Windows download + run the installer from ollama.com/download # Linux curl -fsSL https://ollama.com/install.sh | sh -
Make sure it's running. The macOS/Windows desktop app starts the server
for you (look for the llama in your menu bar / tray). On Linux, or if
you installed CLI-only:
Vizzy expects it on the default port, 11434.ollama serve -
Pull a model — the default Vizzy asks for is
qwen2.5-coder(~4.7 GB):
The model fills in a small JSON patch (not raw shader code), so even small models reliably produce working visuals — change the model name in Vizzy's top bar. Bigger model = better taste, slower drops.ollama pull qwen2.5-coder -
Open Vizzy, type a prompt into a deck, hit
Generate. First generation warms the model up, then it flows.
Builds are fresh off CI and unsigned, because Apple wants £99 a year to
notarize apps and this is a free toy — we ain't paying to notarize nuffink.
So macOS will claim Vizzy "is damaged and can't be opened". It isn't damaged;
Apple is just sulking about the £99. Drag Vizzy.app to Applications, then run
xattr -dr com.apple.quarantine /Applications/Vizzy.app once and
it opens forever after.
This very background is a little plasma shader — Vizzy's native engine
makes this kind of thing all night, only faster and with more knobs.