MiniMax-Music3 explained: the open-weights AI music model
What MiniMax-Music3 is, how the open-weights text-to-music model works, what it can generate, and how OpenTunes uses it to make licensable tracks.
What it is
MiniMax-Music3 is an open-weights text-to-music model: it takes a written description of a track — genre, tempo, key, mood, instruments, vocal style — and optional lyrics, and generates finished, produced audio in 32 kHz stereo. Because the weights are open, it can be run and studied by anyone under its community licence, which allows commercial use subject to conditions such as displaying the model name.
How the generation works
The model reads two things: a structured caption and the lyrics. The richest results come from a detailed, producer-style caption that describes how the sound changes across sections — how the verses feel, what happens at the chorus, how the arrangement builds. Lyrics use section tags on their own lines; wordless tags produce an instrumental.
OpenTunes builds these captions for you. You can write a short idea and let the AI composer expand it into a full structured caption, or fill in the fields yourself in the studio.
How OpenTunes uses it
OpenTunes runs MiniMax-Music3 so you can generate full tracks in the browser for free, then license the ones you want. The catalogue of licence-ready tracks is generated with the same model. Every track carries C2PA provenance identifying it as AI-generated.
Frequently asked questions
What is MiniMax-Music3?
MiniMax-Music3 is an open-weights AI music-generation model that turns a text description (a caption) plus optional lyrics into a full, produced song with vocals or instrumental audio. It is released by MiniMax under a community licence that permits commercial use with conditions.
Can MiniMax-Music3 generate vocals and lyrics?
Yes. It conditions on a caption describing the sound and on lyrics with section tags (intro, verse, chorus, bridge, outro). Leave the lyrics as wordless structure and it produces an instrumental track instead.
How long can generated songs be?
The model can produce multi-minute tracks. In practice the length is driven by how much structure and lyric you give it, up to a per-request ceiling; the duration you request is an upper bound.