Skip to main content
MiniMax Music 3 MiniMax Music 3 is MiniMax’s music generation model for creating complete songs up to five minutes long. Conditioned on lyrics and a detailed music description, it produces structurally coherent songs with expressive vocals, evolving arrangements, and stable long-form audio quality. You can explore generated examples on the official demo page. MiniMax Music 3 uses a hierarchical autoregressive architecture with an 8B global LLM for long-range musical structure, a 0.6B local LLM for frame-level acoustic detail, and a continuous hidden-state synthesis system based on Flow Matching and a Flow-VAE. It outputs 32 kHz, 16-bit stereo audio. The model weights are available under the MiniMax-Music3 Community License. ComfyUI natively supports MiniMax Music 3 with an official example workflow: MiniMax Music 3 Text to Music.
Make sure your ComfyUI is updated.Workflows in this guide can be found in the Workflow Templates. If you can’t find them in the template, your ComfyUI may be outdated.If nodes are missing when loading a workflow, possible reasons:
  1. You are not using the latest ComfyUI version (Nightly version)
  2. Some nodes failed to import at startup

​
Key features

  • Complete songs: Generate full tracks up to 5 minutes with intro, verse, pre-chorus, chorus, bridge, instrumental break, and outro
  • Two-input control: A structured Caption defines the music style, mood, vocals, and arrangement, while Lyrics with section tags ([Intro], [Verse], [Chorus], [Bridge], [Instrumental], [Outro]) control the song structure
  • Long-range coherence: Maintains musical themes, rhythm, vocal identity, and arrangement progression across long sequences
  • Expressive vocals: Natural vocal synthesis with control over melody, pronunciation, and layered harmonies
  • Open weights: Run locally in ComfyUI with full control over every parameter

​
Getting started

MiniMax Music 3 is supported in ComfyUI with open weights. To get started:
  1. Update ComfyUI to the latest version
  2. Go to Template Library > Audio > choose the MiniMax Music 3 workflow
  3. Follow the pop-up to download models and run the workflow
The ComfyUI-ready model files (repacked diffusion model, text encoder, and VAE) are hosted on Hugging Face in the Comfy-Org/MiniMax-Music-3 repository. The original model weights are available from the MiniMaxAI/MiniMax-Music3 repository.

​
MiniMax Music 3 Text to Music workflow

Generate a complete song from a structured music caption and lyrics. MiniMax Music 3 Text to Music workflow preview

Download Workflow

Download JSON or search “MiniMax Music 3” in Template Library
The workflow takes two text inputs and produces a finished song. The text encoder generates the acoustic conditioning sequence from the caption and lyrics, the diffusion model synthesizes the audio latent, and the audio VAE decodes it into a 32 kHz stereo track saved as an MP3. Workflow controls:
  • Caption: The music description. Write it in three sections: Global Metadata (genre, BPM, key, scale, emotional progression, listening scenario, production profile), Vocal Details (vocal gender, timbre, performance style, harmonies, vocal effects), and Arrangement (primary and secondary instruments, groove, bass, percussion, textures, spatial effects). The more specific, the closer the result
  • Lyrics: The words to be sung, with section tags such as [Intro], [Verse], [Chorus], [Bridge], [Instrumental], and [Outro]. Tags are the structural instructions; the lyric text conveys the mood
  • max_duration: Target song length in seconds (the template defaults to 60 seconds; the model supports up to about 300 seconds / 5 minutes). Longer songs take more time and VRAM
  • seed: Random seed for the generation. Keep it fixed to reproduce the same song; change it to get a different take
  • tiled_decode: Decode the audio VAE in overlapping tiles to cut VRAM usage, helpful for long songs on low-VRAM GPUs. Slightly slower with a small risk of seams at tile boundaries. Turn it off on high-VRAM GPUs for best quality

​
Model downloads

Diffusion Model (FP16)

minimax_music3_dit_fp16.safetensors → ComfyUI/models/diffusion_models/

Diffusion Model (INT8)

minimax_music3_dit_int8_convrot.safetensors → ComfyUI/models/diffusion_models/ (for low VRAM)

Text Encoder (INT8)

minimax_music3_text_encoder_pruned_int8_convrot.safetensors → ComfyUI/models/text_encoders/

VAE

minimax_music3_dav.safetensors → ComfyUI/models/vae/
Place the files in the following directories:

​
Prompting tips

  1. Structure the caption in three sections: Global Metadata, Vocal Details, and Arrangement. The model follows not just a global style but the musical development of the song over time
  2. Use section tags in the lyrics: [Intro], [Verse], [Pre-Chorus], [Chorus], [Post-Chorus], [Bridge], [Instrumental], [Solo], and [Outro] give the model explicit structural control
  3. Keep tags as the only structural instructions: The tags are executable; the lyric text itself conveys mood, not structure
  4. Balance duration and resources: Longer songs take more time and VRAM. Use the INT8 diffusion model and tiled decode for long songs on low-VRAM GPUs

​
Prompt writing guide

The official demo page showcases 26 example tracks across 11 genres (pop, rock, R&B, hip-hop, dance pop, country, soul, funk, blues, bossa nova, and urban), each with its full structured caption written in the Global Metadata / Vocal Details / Arrangement format. Use them as reference for writing your own captions. MiniMax also publishes an official music caption rewriter skill that turns a brief music description and optional tagged lyrics into a detailed Music 3.0 structured caption. The skill builds a coherent section-by-section arrangement with global metadata, vocal details, and arrangement descriptions, preserving explicit musical constraints while keeping the lyric text in the lyrics input. Install it with:

​
Output

The generated song is saved to ComfyUI/output/audio/audio_minimax_music3.mp3 (or the filename set in the SaveAudioAdvanced node).