How MiniMax Music takes direction
The MiniMax music models take a compact prompt covering genre, mood, instrumentation and vocal, a lyrics field with [Intro]/[Verse]/[Chorus] markers, and an explicit instrumental flag. The weights are published openly, so the same brief works in the hosted product and in a local deployment — which is why this format is plain text with no vendor-specific field names.
Layer by layer
- The prompt is a summary, not a brief. Genre, mood, two or three instruments, vocal type. Everything else belongs in the lyrics.
- Lyrics carry the detail. Line length, rhyme scheme, where the hook repeats — that is the arrangement, and it is what the model follows.
- Section tags are not optional in practice. Untagged lyrics drift into one long verse shape; tagged lyrics get a chorus.
- Keep the flag and the lyrics consistent. Instrumental on with words in the box asks the model to resolve your contradiction for you, and it will pick one.
- Plan for a normal song length. These models are built for two to three minutes. A six-minute structure is a lot of song for one generation.
- Name the language if it is not English. The models are multilingual and will otherwise infer from the prompt text.
MiniMax Music setups you can paste
Keep the prompt to one line, put the detail in the lyrics, and set the flags to match.
Neo-soul · 88 BPM · B minor
Festival house · 126 BPM · F minor
Documentary bed · 90 BPM · instrumental
More ready-made prompts across moods and engines: AI music prompt examples.
What to leave out
- Essays in the prompt. The field is short by design; the lyrics are where detail pays off.
- Untagged lyrics. Without markers there is no chorus, just a wall of lines.
- Flag and lyrics in conflict. Decide instrumental or vocal before you generate, not after.
- Expecting a negative prompt. Fold avoidances into the wording: name what you want present.
- Artist names. Rights risk, weaker brief, no upside.
The rule underneath all of them: describe the music, never the record. Reference an artist or a track and you inherit someone else's rights problem along with a vaguer brief.
Build a MiniMax Music brief from a reference track
The hard part is the honest numbers: tempo, key, brightness, dynamics, and the instruments actually carrying the groove. Music DNA Studio measures them from any MP3, WAV, FLAC or MP4 — locally in your browser, nothing uploaded — then outputs a MiniMax Music-shaped prompt alongside versions for the other ten engines. You make the creative calls; the spec underneath is measured, not guessed.
Because the format is plain text, a measured reference transfers cleanly: tempo, key, brightness, dynamics and instrumentation, written as one compact line plus the lyric structure you actually want.
Guides for the other engines
Eleven Music · Suno · Udio · Mureka · MiniMax Music · YuE2 · Stable Audio · Soundraw · Mubert · Beatoven · AIVA
Build The MiniMax Brief
Drop in a reference, pick MiniMax Music, copy the prompt, lyrics and flags. Free, no signup, nothing leaves your device.
ANALYSE YOUR AUDIO — FREE →FAQ
How long should the prompt be? ▾
One line. Genre, mood, two or three instruments and the vocal type is enough — the field is deliberately compact, and pushing a paragraph into it crowds out the lyrics, which carry more weight.
How do I get an instrumental track? ▾
Turn the instrumental flag on and leave the lyrics empty. Leaving words in the box while the flag says instrumental gives the model a contradiction to resolve, and you will not like its choice.
What does "open weights" mean for me? ▾
The model can be run outside the vendor's website — locally on your own hardware or in another hosted service. The brief stays the same; check the licence before building a product on top of it.
Why do the vocals sound flat? ▾
Usually no vocal description in the prompt, plus lyrics with no dynamic contrast between sections. Name the voice and give the bridge genuinely different lines from the chorus.
Which languages are supported? ▾
Several, and the models infer language from the text. If you are writing in a language other than English, state it in the prompt as well as writing the lyrics in it.