Skip to content

MiniMax Music 3.0 Studio

MiniMax Music 3.0 Studio is a local-first generation workspace for ComfyUI. It turns a structured brief and section-tagged lyrics into a real queue job, then keeps the resulting MP3 available in a small local Library.

What it provides

  • Compose — separate fields for genre, tempo, key, vocal direction, arrangement, and tagged lyrics.
  • Local gateway — a dependency-light Python bridge between the browser and the checked-in MiniMax Music 3.0 workflow.
  • Library — local task metadata plus browser playback and MP3 download.
  • Setup — visible ComfyUI and gateway health with copyable Windows commands.

The UI is the primary product surface. SRT generation and HyperFrames lyric motion are intentionally separate downstream workflows that consume a Library MP3.

Quick start

  1. Start ComfyUI with its API on 127.0.0.1:8201.

  2. From the repository root, start the gateway:

    powershell
    python tools/music3_gateway.py
  3. Start the frontend:

    powershell
    Set-Location frontend
    npm ci
    npm run dev
  4. Open http://127.0.0.1:5173 and choose a Quick start in Compose.

Set MUSIC3_COMFY_URL, MUSIC3_GATEWAY_PORT, or VITE_API_TARGET when the local services use different addresses. Model weights are not included in the repository.

Request flow

text
Compose form
    │ structured caption + tagged lyrics

MiniMax Music 3.0 gateway :8202
    │ POST /prompt

ComfyUI :8201 ── GPU render ──► SaveAudioMP3
    │ history polling

Library metadata + /audio/<prompt-id>

The checked-in workflow uses the INT8 DiT, MiniMax text encoder, MiniMax DAV VAE, KSampler, and SaveAudioMP3. Tiled decode selects VAEDecodeAudioTiled; regular decode uses VAEDecodeAudio.

Verification

From the repository root:

powershell
docker compose -f docker-compose.music3.yml config --quiet
Set-Location frontend
npm run typecheck
npm test
npm run build
Set-Location ..
python -m unittest tools.test_music3_gateway

The practical smoke path is: open Compose, choose a Quick start, use a short duration, generate, wait for Ready to audition, and play the MP3 from the queue or Library.

Boundaries

  • Keep model weights, private .env files, audio, SRT, and rendered video out of Git.
  • The public repository is an unofficial community project and is not affiliated with MiniMax, ComfyUI, or the ACE-Step team.
  • The React shell started from ACE-Step Forge; the MiniMax gateway, workflow, and Studio UI changes live in this repository.

See the Japanese guide or return to the repository README.

MiniMax Music 3.0 Studio is an unofficial community project.