MiniMax Music 3: open-weights songs from your lyrics on Mage

MiniMax Music 3 is MiniMax’s open-weights music model. You write the lyrics and describe the music, and it performs a complete song up to five minutes long: verses, choruses, a bridge, expressive vocals, and an arrangement that develops from the intro to the outro. It also makes instrumentals. Mage runs it on its own GPUs and returns a 44.1 kHz stereo MP3. It takes MiniMax Music 3 LoRAs.

Overview

Detail
Type	Audio
Origin	Open-weights model hosted on Mage’s GPUs
Models	MiniMax Music 3
Unlimited from	Basic

Unlimited generation by plan

MiniMax Music 3 is priced by the Length you pick, the longest the song may run.

Model	Setting	Basic	Pro	Pro Plus	Max
MiniMax Music 3	30 or 60 seconds	Unlimited	Unlimited	Unlimited	Unlimited
MiniMax Music 3	120 or 180 seconds	Gems	Unlimited	Unlimited	Unlimited
MiniMax Music 3	240 or 300 seconds	Gems	Gems	Unlimited	Unlimited

What it supports on Mage

Feature	Support
Your own lyrics	Yes, in the Lyrics setting
Instrumentals	Yes: leave Lyrics empty or turn on Instrumental
Lengths	Up to 30, 60, 120, 180, 240, or 300 seconds
LoRAs	Yes
Seed	Yes
Audio references	No
Characters and References	No

Best for

Full songs that sing your own lyrics, word for word
Song structure in the order you tag it: verses, choruses, bridges, solos, outros
Arrangements that build, with instruments entering and dropping out by section
Instrumental tracks and scores
Songs in English and Mandarin Chinese

Writing prompts

MiniMax Music 3 takes two inputs. The prompt describes the music: the genre and subgenre, the tempo in BPM, the key, the mood and how it builds, the singer, and what the instruments do in each section. Lyrics, on the Advanced page, holds the words to sing. Put each section tag ([intro], [verse], [pre-chorus], [chorus], [bridge], [instrumental], [solo], [outro]) alone on its own line, with the sung lines below it. Words on the same line as a tag are dropped. Always describe the singer, such as “warm female lead, breathy in the verses, soaring in the chorus”, or the song can come out instrumental. Length is a limit, not a target: the model ends the song when the lyrics and the arrangement are done, and stops at the limit even mid-line. A verse and a chorus fill about a minute. See the prompting guide for templates and examples.

Example prompt and lyrics to start from

The prompt:

Indie folk pop ballad, 84 BPM, G major. Tender and hopeful, opening hushed and swelling into a wide, glowing final chorus. Vocals: warm female lead, close and breathy in the verses, opening up with soft stacked harmonies in the chorus. Arrangement: fingerpicked acoustic guitar and felt piano carry the verses; cello and brushed drums enter at the first chorus; a gentle string swell lifts the last chorus, then the song falls back to guitar alone.

The lyrics:

[verse]
We wrote our wishes on paper lanterns
Down by the river when the night was new
[chorus]
So let them rise, let them rise
Little fires in a velvet sky
[outro]
I will light the way

Using LoRAs

Pick a MiniMax Music 3 LoRA under Model → LoRA, or bring your own on the Import page by choosing MiniMax Music 3 as its model. LoRAs import from Hugging Face or CivitAI; checkpoints do not. See LoRAs for how weights and stacking work.

Limitations

It sings the lyrics you write; it does not write lyrics for you. For a song whose lyrics the model writes, use Eleven Music.
No audio references: it cannot copy a voice, a melody, or a recording.
Tempo, key, and structure are followed as guidance, not exactly.
English and Mandarin Chinese are the languages MiniMax demonstrates; others are untested.

Use via API

MiniMax Music 3 is available through the Mage API as the minimax_music architecture, with the words to sing in lyrics. Its fields, tokens, defaults, and price are on its API reference page.

Models

Mage models

Open-weights models

Models run by their lab

Audio