MiniMax Music 3 - ComfyUI Song Generator Workflow (Local AI Music with Vocals)

A ready-to-run ComfyUI workflow for MiniMax Music 3, the new open-weights AI music model — one of the best local Suno alternatives out right now. Feed it lyrics and a music description and it generates a complete song with vocals, instruments, and full structure, up to 5 minutes, in 32 kHz stereo. Runs locally, no subscription, and light enough for ~8GB VRAM territory.

The example workflow is attached to this post — drag the .json into ComfyUI and load your three models.

Required Models

Update ComfyUI to 0.33.0+ first. Three files from the Comfy-Org/MiniMax-Music-3 repo:

• minimax_music3_dit_fp16.safetensors → models/unet/ (I recommend FP16 over the INT8 for quality; INT8 exists if you're tight on space)

• minimax_music3_text_encoder_pruned_int8_convrot.safetensors → models/clip/

• minimax_music3_dav.safetensors → models/vae/

If the model doesn't show in the node dropdowns, a file is in the wrong folder — that's the usual cause.

Writing the Caption

Output quality lives in the caption. Write it in three parts — Global metadata (genre, BPM, key, mood, production), Vocal details (gender, tone, delivery), and Arrangement (how sections evolve) — and tag your lyrics with [Verse] / [Chorus] on their own lines. Easiest approach right now is to have an LLM format your idea into that structure while keeping your lyrics intact.

Notes

• FP16 for final renders — noticeably better than INT8 if you have the VRAM.

• Keep first generations short to dial in the caption before committing to a full track.

• Generation time scales with length and GPU.

• Licensing: MiniMax Music 3 uses MiniMax's community license (not fully open). Free for personal use; if you monetize, you're asked to credit "MiniMax-Music3" and disclose AI-generated audio. Full terms on the HF page.

More From Me

If you're into local AI video too, check out my MiniMax H3 work — the sister model that does text/image to video with native audio:

• MiniMax H3 ComfyUI Workflow + Turbo LoRAs (4-8 step speedup, native audio) — on my Patreon: https://www.patreon.com/TheLocalLab

• MiniMax H3 RunPod Template (run it in the cloud, no local GPU): https://get.runpod.io/Minimax-H3-ComfyUI

More free workflows, one-click installers, and tutorials: ▶️ YouTube - https://www.youtube.com/@TheLocalLab 🎨 Patreon - https://www.patreon.com/TheLocalLab 🛒 https://www.locallabdigest.com

If this workflow helped, a ❤️ or review is appreciated!