MiniMax Music 3
MiniMax Music 3WorkflowComfyUIAI Music SOP

MiniMax Music 3 Workflow Guide: Prompting SOP, ComfyUI & Music Pipeline

Master the complete MiniMax Music 3 workflow. Learn how to write structured captions, format lyric tags ([Verse], [Chorus]), setup ComfyUI nodes, and generate 5-minute full songs.

Create AI Music for FreeExplore more AI music guides
MiniMax Music 3 Workflow Guide: Prompting SOP, ComfyUI & Music Pipeline
Last UpdatedAug 16, 2026
Category
Tutorials
Summarize withChatGPTPerplexityGeminiClaude

The release of MiniMax Music 3 (Music 03) marks a major shift in AI music generation. Unlike legacy text-to-audio models that produce brief 30-second clips or lack structural coherence, MiniMax Music 3 is designed to generate complete, high-fidelity songs up to 5 minutes in duration with 32kHz 16-bit stereo quality.

However, getting pro-grade results requires understanding the MiniMax Music 3 Workflow. In this guide, we break down the exact workflow—from prompting and lyric section tagging to ComfyUI node setups and post-production.

1. The Dual-Control Architecture Workflow

To get the most out of MiniMax Music 3, you must understand how its hybrid 8B Global LLM + 0.6B Local LLM system interprets input. The model relies on a two-part input pipeline:

  1. Structured Caption: Defines global metadata, instrumentation, vocal timber, BPM, and acoustic progression.
  2. Track Lyrics with Section Tags: Guides the structural arc using standardized tags like [Intro], [Verse], [Chorus], and [Outro].
[User Input] ──► Structured Caption + Tagged Lyrics
                      │
                      ▼
[8B Global LLM] ──► Plans Melody, Phrasing & Structural Blueprint
                      │
                      ▼
[0.6B Local LLM] ──► Fills Frame-level Acoustic & Timbre Details
                      │
                      ▼
[Flow VAE] ──────► Renders 32kHz 16-bit Stereo Audio Output (.wav)

2. Prompting & Song Architecture SOP

Step 1: Crafting the Structured Caption

Avoid short, generic prompts like "happy pop song". MiniMax Music 3 performs best when provided with detailed metadata blocks.

[Genre] Modern Synthwave / Cyberpunk Electronic
[BPM] 120 BPM
[Key] A minor
[Vocals] Female lead, ethereal voice with vocal harmomies in chorus
[Instruments] Analog bass synth, gated drums, arpeggiated lead synth, warm pads
[Arrangement] Atmospheric quiet intro, building energy into a punchy chorus, melodic guitar solo in bridge

Step 2: Formatting Lyrics with Structural Tags

Each section tag must reside on its own dedicated line. Do not put lyrics on the same line as a bracketed tag.

Supported Tags:

  • [Intro]: Instrumental introduction or spoken intro.
  • [Verse] / [Verse 1]: Verse sections establishing narrative flow.
  • [Pre-Chorus]: Building momentum toward the hook.
  • [Chorus]: The main emotional and melodic climax.
  • [Post-Chorus]: Rhythmic or vocal hook following chorus.
  • [Bridge]: Harmonic change or transition section.
  • [Instrumental] / [Solo]: Guitar, synth, or saxophone solo without vocals.
  • [Outro]: Fading outro or final resolving chords.

Example Lyric Block:

[Intro]
(Atmospheric synth pads swelling slowly)

[Verse 1]
Neon lights reflect in rain
Digital dreams inside my brain
Walking through the silent street
Counting every electric beat

[Pre-Chorus]
System boot, sequence start
Binary blood inside my heart

[Chorus]
We run through the midnight glow
Where the synthwave rivers flow
Singing louder than the thunder
Falling in a neon wonder

[Outro]
Signal fades... fading out...
(Soft delay synth echo)

3. ComfyUI Local Workflow Setup

For creators running MiniMax Music 3 locally on consumer GPUs (recommended: 16GB+ VRAM, minimum 8GB with layer offloading):

1. Download Model Weights

Download the model checkpoint from Hugging Face:

git clone https://huggingface.co/MiniMaxAI/MiniMax-Music3

2. Download Ready-to-Use Workflow JSON

We have prepared a pre-configured ComfyUI workflow template for MiniMax Music 3:

📥 Download MiniMax Music 3 ComfyUI Workflow (.json)

3. Load into ComfyUI

  1. Drag and drop minimax-music-3-workflow.json into your ComfyUI workspace.
  2. Select your MiniMaxMusic3Loader node and set offloading to group_offloading if your GPU has 8–12GB VRAM.
  3. Paste your Structured Caption and Tagged Lyrics into the respective prompt nodes.
  4. Click Queue Prompt to render.

4. One-Click Cloud Web Workflow (No GPU Required)

If you do not have a high-end GPU or prefer not to configure ComfyUI locally, you can use the instant Web Workflow on MiniMax Music 3 Web:

  1. Navigate to the generator studio on the home page.
  2. Select a preset workflow (e.g., Pop, EDM, Cinematic, Lo-Fi).
  3. Customize your lyrics or click AI Enhance to automatically structure your prompt.
  4. Click Generate Music to receive a full 5-minute track in seconds.

5. Post-Production & DAW Stems Pipeline

Once you have generated your track, follow this DAW workflow for commercial production:

  1. Stem Separation: Use AI stem splitters (like Ultimate Vocal Remover or Lalal.ai) to isolate Vocals, Drums, Bass, and Synth tracks.
  2. EQ & De-Essing: Apply a high-pass filter on vocals around 80Hz and de-ess sibilant frequencies.
  3. Mastering Limiter: MiniMax Music 3 outputs 32kHz stereo WAV; apply a subtle brickwall limiter (-0.3 dBFS peak) in Ableton or FL Studio to maximize loudness for streaming platforms.

Summary Checklist

  • Use Structured Caption format [Genre] [BPM] [Key] [Vocals] [Instruments].
  • Put lyric tags ([Chorus], [Verse]) on separate lines.
  • Use minimaxmusic3.app for instant Web generation or download our ComfyUI Workflow JSON for local setups.