Minimax Music 3 is now on ComfyUI

Conditioned on lyrics plus a music description, it holds theme, rhythm, vocal identity, and arrangement together across a full five-minute track and outputs 32kHz 16-bit stereo. The reason it holds up: An 8B Global LLM tracks long-range structure while a 0.6B Local LLM restores frame-level acoustic detail, and the synthesis stage fuses both hidden states through Flow Matching and a Flow-VAE decoder rather than decoding from discrete tokens. Continuous representations carry more acoustic information than tokens can you hear it in vocal articulation and instrumental texture. You control the structure with section tags in the lyrics - [Verse], [Pre-Chorus], [Chorus], [Bridge], [Solo], [Outro] - and steer style with a Structured Caption covering global metadata, vocal details, and arrangement evolution. Download the workflow: https://links.comfy.org/45qDiII