Which model variant should I use: fp16 or int8?

Use minimax_music3_dit_fp16.safetensors if you have enough GPU VRAM for best fidelity and stability. Choose minimax_music3_dit_int8_convrot.safetensors when VRAM is limited; it’s more memory-efficient with a potential slight quality trade-off.

How should I structure the Caption and Lyrics for best results?

Keep the Caption explicit: genre, mood, tempo/BPM, key instruments, production style, and arrangement (e.g., intro–verse–chorus–bridge–outro). For Lyrics, include clear section tags like [intro], [verse], [chorus], [bridge], [outro]. Keep lines concise and align section lengths with your requested duration.

Can it generate vocals or is it instrumental only?

If you provide Lyrics with section tags, the model will attempt to render vocal-like parts following your text and structure. If you leave Lyrics empty and supply only a Caption, it will generate instrumental music.

What is the maximum song length and how do I set it?

The workflow supports up to about 5 minutes (300 seconds). Set the duration parameter on the MiniMax Music 3 generator node and keep your arrangement (in Caption/Lyrics) sized to fit that length.

MiniMax Music 3: Text to Music

This ComfyUI tutorial workflow turns structured text into full-length music using MiniMax Music 3. At its core is the MiniMax Music 3 generator node (UUID: ac99f841-a3de-4329-9564-953b81cf9e16), which accepts two key inputs: a Caption describing style, mood, tempo, instrumentation, and arrangement, and optional Lyrics formatted with section tags like [intro], [verse], [chorus], [bridge], and [outro]. The node renders a complete track—up to five minutes—with stable song structure and high-fidelity audio. A MarkdownNote block in the graph documents the prompt schema so you can quickly copy, tweak, and version your captions and lyrics inside the UI.

Under the hood, you can choose between two model weights: minimax_music3_dit_fp16.safetensors (recommended when VRAM allows) and minimax_music3_dit_int8_convrot.safetensors (optimized for lower VRAM). The generator node produces an audio tensor that is written to disk via SaveAudioAdvanced, where you control filename, format, and sample rate. The result is a finished, arranged song suitable for prototypes, background scores, and content production without leaving ComfyUI.

API

Use this workflow from code

Every Comfy workflow is a JSON graph. The payload below is this workflow, exactly as ComfyUI runs it — fetch it from the URL, keep it in version control, or load it in ComfyUI and run it node by node.

04479f67187d.json

Fetching workflow JSON…

Run it from TypeScript or Python with the Comfy SDK. The same code targets Comfy Cloud or a ComfyUI you host yourself — only the base URL changes.

// Install (beta)
npm i @comfyorg/sdk

// Run this workflow (TypeScript)
import { Comfy } from "@comfyorg/sdk";

const client = new Comfy({ apiKey: "comfyui-..." });
const wf = await client.workflows.fromFile("workflow_api.json");
const job = await client.run(wf);
await job.getOutputs("<output-node-id>")[0].toFile("output.png");

The SDK takes a workflow in API format: open this workflow in ComfyUI and use File → Export Workflow (API).

Comfy Cloud API access requires a plan with an API key.

SDK docs Get an API key

FAQ

Frequently Asked Questions