MiniMax Music 3: State of the Art Open Weight Music Generation

Full songs up to five minutes, driven by lyrics and a music description — expressive vocals, evolving arrangements, and 32 kHz stereo output.

Another big day for the open source community! A new state of the art open weights music generation model has arrived in ComfyUI. MiniMax has released MiniMax Music 3. Give it lyrics and a description of the sound you want, and it returns a complete, structurally coherent song up to five minutes, in 32 kHz stereo.

MiniMax Music 3 is built for complete songs rather than loops or clips: conditioned on lyrics and a music description, it holds themes, rhythm, vocal identity, and arrangement progression together across the full track, and outputs 32 kHz, 16-bit stereo audio.

The architecture is a hybrid of two language models and a continuous synthesis stage. An 8B Global LLM (initialized from Qwen3-8B) models the song’s long-range semantic and structural progression. A 0.6B Local LLM restores fine-grained acoustic detail within each frame. Instead of decoding audio from discrete tokens alone, the synthesis module fuses the hidden states of both LLMs and passes them through a 2.4B Flow Matching stage and a 123M Flow-VAE decoder. Those continuous representations carry more acoustic information than tokens can which shows up in vocal articulation, instrumental texture, and temporal continuity.

Model Highlights

• Full songs with long-range coherence: Native support for generation up to five minutes. The model sustains complete song structures including intro, verse, pre-chorus, chorus, bridge, instrumental break, and outro, without losing the thread on melody, rhythm, or vocal identity.

• Lyrics with section tags: The lyrics input accepts explicit structure tags: [Intro], [Verse], [Pre-Chorus], [Chorus], [Post-Chorus], [Bridge], [Instrumental], [Solo], and [Outro]. You decide the song structure.

• Fine-grained control via Structured Captions — Beyond a plain-language description, the model accepts a Structured Caption with three sections: Global Metadata (genre, BPM, key, scale, emotional progression, production profile), Vocal Details (gender, timbre, performance style, harmony, backing vocals, effects), and Arrangement (instruments, section-level instrument evolution, groove, bass, percussion, textures, spatial effects). This lets you steer not just the overall style but how the song develops over time.

• Expressive vocals — Vocal performance is controllable down to timbre, harmony, and effects, and the hidden-state synthesis path is what keeps articulation clean across long passages.

• 32 kHz stereo output — 16-bit stereo WAV, with the Flow-VAE decoder retrained for the dynamic range and spectral characteristics of music.
  
  MiniMax also published 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:
  npx skills add MiniMax-AI/MiniMax-Music3 --skill music-caption-rewriter

Examples

Gospel Soul

Rock (Instrumental)

Lo-fi Chillhop

Getting Started

1. Update ComfyUI to the latest version 0.33.0 or go to Comfy Cloud.

2. Download the workflow below, or find it in the template library.
   

3. Follow the note in the workflow to download the model weights and save them in the correct model directory.

4. Add your lyrics and music description, then run the workflow.

Model weights: 🤗 MiniMaxAI/MiniMax-Music3

As always, enjoy creating!