MiniMax Music 3 GGUF
The MiniMax Music 3 autoregressive text encoder + flow-matching DiT, converted to GGUF.
Complete songs with vocals from a caption (music description) + lyrics. Up to ~5 min, stereo 44.1 kHz.
Samples
Synth-Pop - 2:29 min (Q8_0 GGUF)
- ~7.3 min on an RTX 5090 (32 GB), peak 9.7 GiB.
| Parameter | Value |
|---|---|
| Prompt | Polished modern synth-pop with a cinematic lift. 104 BPM, A minor. Bright, wide and euphoric, building from a restrained verse into a soaring chorus. High-fidelity studio production: pristine and clean, crisp transients, wide stereo image, deep controlled sub, glossy modern mastering. No vinyl noise, no tape hiss, no lo-fi texture. Vocal Details: Clear, confident female lead, forward and centered in the mix, pop-clean and articulate with strong sustained notes in the chorus. Bright airy top end, tight controlled vibrato. Stacked octave harmonies and a layered choir-like double on the chorus, wide and shimmering. |
Punk-funk — 1:01 (Q8_0 GGUF)
- ~5 min on an RTX 5090 (32 GB), peak 9.7 GiB.
| Parameter | Value |
|---|---|
| Prompt | Modern pop-punk, fast and bright four-piece band, in the vein of late-nineties California skate punk. 172 BPM, E major, straight eighths, relentless forward drive. Energetic, sunny and a little bratty, sing-along and radio-ready. High-fidelity modern rock production: tight and punchy, crisp cymbals, clear midrange, powerful but controlled low end, polished commercial master. No lo-fi texture, no vinyl noise, no tape hiss. Snotty melodic male lead with a bright nasal edge and a slight sneer, sung not screamed, forward and dry in the center with a short slapback. Big shouted gang-vocal "whoa-oh" backups stacked wide across the chorus, and a second harmony vocal a third above the lead on the hook. |
Components used to generate
| Component | File | ~Size | Download |
|---|---|---|---|
| Text Encoder | minimax_music3_text_encoder_pruned_Q8_0.gguf | 8.88 GB | Link |
| Transformer | minimax_music3_dit_Q8_0.gguf | 2.70 GB | Link |
| Audio VAE | minimax_music3_dav.safetensors | 217 MB | Link |
⚠ Must use the updated GGUF Loader ⚠
In comfy:
- Open the ComfyUI Manager
- Change the channel to "Channel (remote)"
- and search for comfyui-gguf-loader
Use the workflow Link
Caption in three parts:
- global metadata (genre, BPM, key, mood)
- vocal details
- arrangement.
Note: Describe the vocal explicitly (ie. "warm female vocal, close and breathy" etc.) or it drifts.
Structure tags each on their own line:
[intro]
[verse]
[pre-chorus]
[chorus]
[bridge]
[instrumental]
[solo]
[outro].
Note: These are the only tags the model uses.
Recommended settings
| Parameter | Value | Notes |
|---|---|---|
duration | 120–300 | 300s Max |
steps | 30 | Euler scheduler |
guidance /CFG | 1.7 | flow stage |
Sources
| Base model | MiniMaxAI/MiniMax-Music3 |
| Source weights | Comfy-Org/MiniMax-Music-3 |