What “AI music visualizer” means here
Most things sold as AI video tools render one fixed clip from a prompt. That clip does not listen to anything. A visualizer is the opposite: a program that takes the music as input and draws something new every frame from the bass, mids, treble, and beats. So the useful thing for AI to generate is not a video but the program, and that is what Groovalizer does. The output is a preset in the MilkDrop .milk format, the same format behind the 3,500+ presets already in the library, which means it plays live, exports to MP4, and can be edited by hand afterwards.
Because the preset runs on the real projectM engine, there is no faking it: a preset either renders and moves with the song, or it fails and says why. That is also what makes revising possible.
Three ways to do it
1. Any chatbot, with the prompt Groovalizer gives you
Free
Open Import in the studio and copy the AI prompt. It contains the preset format, a worked example, and the rules that keep motion coherent. Paste it into ChatGPT, Claude, Gemini, or a local model, describe the look you want, and paste the .milk code it returns back into Import. It loads on the real engine immediately.
Best for: Anyone with a chatbot subscription they already pay for, and no interest in setup.
2. Your own coding agent, connected over MCP
Free: you bring the agent
Groovalizer exposes a remote MCP (Model Context Protocol) server. Connect Claude Code or any MCP client with a token from your account, and the agent can write a preset, load it in your live studio tab, read the engine’s errors, capture frames of what it rendered, and refine: a real feedback loop rather than a single guess.
Best for: Developers and anyone who already works with an agent and wants it to iterate until the visual is right.
3. The built-in agent
Metered in credits, included with Pro or a top-up pack
Open the Agent console, describe the visual, and Groovalizer’s managed agent runs the same loop for you (authoring, loading, checking the result on the engine, and revising) without you connecting anything. Runs are billed on what the model actually consumed.
Best for: People who want to type a sentence and come back to a preset.
What to describe
The prompt Groovalizer supplies already covers the technical rules, so your part is the picture. The descriptions that work best name a subject, a palette, a pace, and what should happen on the beat:
- “A slow violet-and-gold nebula that breathes with the bass and flares white on the kick.”
- “Liquid chrome ribbons folding over each other, cool blues, faster on the hi-hats.”
- “A calm mandala of six petals, pastel, that changes hue every bar.”
Avoid asking for text, logos, or a specific photo: a preset is maths and shaders, not a compositor. Upload an image as a texture if you want your artwork inside the scene.
After it is generated
A generated preset is a normal preset. Save it to your library, tune speed and reactivity with the vibe controls, put it on the timeline for the chorus, run it live through a microphone, or export it as a 9:16 clip for TikTok or a full-length video for YouTube at up to 4K.
And because it is plain .milk code, you can open it in the editor and change a number to see what happens. That is how most people learn the format.