How to Use MiniMax Music 3: Prompt, Lyrics, Structure

I opened the MiniMax Music 3 generator and found two boxes: Music brief and Lyrics. That split looks simple, but it answers the question that causes most weak first attempts. The brief should describe how the record sounds and develops. The lyrics should tell the model what happens in each section.

If you put a plot summary in the brief and a wall of unlabeled lines in the lyrics box, the model has to invent the arrangement. Give each box one job and the result becomes much easier to steer.

MiniMax describes Music 3.0 as a model that composes, arranges, performs, and produces a complete song from a concept plus optional lyrics. Its official release article emphasizes three upgrades: closer interpretation of creative intent, more complete long-form arrangements, and clearer instruments and vocals. This tutorial turns those ideas into a repeatable workflow. It explains the controls on this site; as the official model card limitations note, section tags and descriptions provide generative control rather than strict guarantees.

The short formula

Use this order for the music brief:

GENRE + TEMPO/KEY + EMOTIONAL ARC + VOCAL + CORE INSTRUMENTS
+ SECTION CHANGES + PRODUCTION CHARACTER

Then give the lyrics their own map:

[intro]
[verse]
[pre-chorus]
[chorus]
[verse]
[chorus]
[bridge]
[chorus]
[outro]

You do not need every field or every section. The formula is a checklist, not a rigid syntax. A two-minute instrumental can work with genre, mood, instruments, and an arrangement arc. A five-minute vocal track needs more explicit structure because there is more time for the idea to drift.

1. Start with a production brief, not a topic

“A song about driving at night” gives the model a subject. It says almost nothing about the record.

Write the sonic decisions a producer would make:

Cinematic synth-pop at 108 BPM, reflective verses growing into a wide
euphoric chorus, intimate female alto lead, pulsing analog bass, crisp
electronic drums, glassy arpeggios, warm strings entering in the
pre-chorus, and a spacious polished night-drive mix.

That brief contains seven useful decisions:

1. Genre: cinematic synth-pop.
2. Tempo: 108 BPM.
3. Emotional movement: reflective to euphoric.
4. Lead voice: intimate female alto.
5. Rhythm section: analog bass and electronic drums.
6. Arrangement event: strings enter in the pre-chorus.
7. Mix target: spacious, polished, night-drive character.

The last two matter more than another stack of genre adjectives. MiniMax's Structured Caption approach is designed to follow music over time, including emotion, instrument entrances, vocal delivery, and production character. Tell it what changes and where.

2. Describe an arc instead of piling up moods

Prompts often fail because they ask for contradictory feelings without assigning them to sections:

sad, joyful, intimate, huge, calm, energetic

Turn that list into a sequence:

Restrained and reflective in the verses, warmer in the pre-chorus,
then open into a euphoric chorus. Pull the drums back for a quiet bridge
before the final chorus returns with wider harmonies and added strings.

Now the contrast has a timeline. The model can treat the quiet bridge as a deliberate reset instead of an instruction that conflicts with “energetic.”

For long songs, include two or three arrangement events. Useful events include:

• bass enters after the first verse;
• harmony vocals appear only in the chorus;
• drums drop out for the bridge;
• a guitar, piano, or synth solo replaces a vocal section;
• the outro returns to the opening instrument.

Do not choreograph every second. You are defining the song's landmarks, not writing automation for a DAW.

3. Put structure tags on their own lines

For a vocal track, the lyrics field is required on this generator. Use section labels so the model can see the macrostructure.

[verse]
Streetlights blur into the rain
I count the exits, miss your name

[pre-chorus]
The city falls behind
The signal turns to blue

[chorus]
Every open road comes back to you
Past the midnight signs and morning blue

[bridge]
Let the engine fade
Leave one voice and distant strings

[chorus]
Every open road comes back to you
Now the skyline opens into view

MiniMax's article explicitly lists tags such as [intro], [verse], [pre-chorus], [chorus], [bridge], [instrumental], [solo], and [outro]. Use the sections the song actually needs. A short track with six repeated labels can feel more crowded than structured.

Three practical lyric rules:

1. Keep each tag on its own line.
2. Keep line lengths singable; split prose into phrases.
3. Repeat the chorus text when you want a true recurring hook. Do not assume a label alone will reproduce exact words.

4. Keep sound instructions out of the lyric lines

The lyrics box is not the best place for a paragraph about microphones, compressors, or synth patches. Put those choices in the brief. Keep bracketed lyric tags focused on musical sections.

Weak:

[verse with dark reverb and huge analog synth but soft drums]
I wait beside the window...

Cleaner:

Music brief: Keep the verse intimate with dry lead vocals, a muted pulse,
and distant glassy synths. Open the reverb and add full drums in the chorus.

[verse]
I wait beside the window...

This separation also makes revisions faster. You can change the production without rewriting the song, or replace the lyrics while keeping the same sonic brief.

5. Use Instrumental mode deliberately

Choose Instrumental when you do not want vocals. This generator sends the supported [instrumental] structure tag and removes the lyrics requirement. The music brief then carries all the important decisions.

A useful instrumental brief still needs motion:

Modern cinematic ambient score, 82 BPM, minor key. Begin with felt piano
and low room tone, introduce warm cellos after the first phrase, add a
restrained electronic pulse at the midpoint, then rise into a broad but
not bombastic final minute before ending on solo piano. No vocals.

“Ambient instrumental” is a category. The longer prompt is an arrangement.

6. Choose the shortest duration that fits the idea

The generator offers 30 seconds, 1 minute, 2 minutes, 3 minutes, and 5 minutes. Music 3.0 supports songs up to five minutes, but longer is not automatically better.

• 30 seconds: a hook, bumper, intro, or quick direction test.
• 1 minute: verse-and-chorus sketch.
• 2 minutes: compact complete song or structured instrumental.
• 3 minutes: conventional song with bridge and final chorus.
• 5 minutes: long-form arrangement with enough lyric and musical development to justify it.

On this site, the displayed cost is 4 credits per requested 30 seconds. Failed generations are refunded automatically; unused 30-second blocks are refunded when the returned track ends early. Check the cost beside the duration selector before generating.

Start with 30 or 60 seconds, fix the brief, and only then request a longer track. That avoids paying for five minutes when the arrangement problem is already apparent in a short draft.

7. Run a three-pass revision

Do not rewrite everything after one weak result. Diagnose one layer at a time.

Pass 1: identity

Check genre, tempo, voice, core instruments, and overall production. If the track feels like the wrong record, fix the brief before touching the lyrics.

Pass 2: structure

Check whether the chorus is clearly different from the verse, whether the bridge creates a reset, and whether key instruments enter at the requested point. Fix section tags and arrangement events.

Pass 3: performance detail

Only then refine vocal tone, articulation, harmony behavior, drum character, room size, or mix density. These details cannot rescue a song with the wrong identity or shape.

If you need a broader framework for writing constrained prompts, the same “assign every input one job” principle appears in our MiniMax H3 prompt guide. The medium differs, but the discipline is the same.

Common mistakes

Writing only a genre list

“Pop, rock, R&B, cinematic” makes the model negotiate four directions. Choose a primary genre, then use one secondary influence for rhythm, harmony, or production.

Naming instruments without roles

“Piano, strings, guitar, synth” is inventory. Say when each enters and whether it leads, supports, or decorates.

Asking for a long song with short lyrics

A five-minute request with one verse and one chorus invites repetition or improvised structure. Add enough sections, instrumental space, or repeated hook material for the selected duration.

Using protected lyrics or asking for a living artist's exact sound

Write original lyrics and describe musical traits directly: tempo, register, instrumentation, vocal delivery, arrangement, and mix. That is more controllable and avoids depending on a third party's identity or catalog.

Changing five variables at once

You will not know which change helped. Keep the prompt and downloaded result together as a revision record, then change one layer at a time.

Three steps to make your first track

1. Open the MiniMax Music 3 tool, choose Lyrics or Instrumental, and start with a 30- or 60-second duration.
2. Paste a brief using the seven-part formula. For vocals, add section-tagged original lyrics with one clear chorus.
3. Generate, preview the WAV, note the structural problem you hear, then revise one layer before requesting a longer version.

You can hear the range of the model on MiniMax's official Music 3 demo page. Use those examples to study how detailed briefs connect genre, tempo, vocal character, instrumentation, and a full-song arc—not as promises that copying a prompt will reproduce the same recording.

Bottom line

MiniMax Music 3 works best when the brief and lyrics behave like two parts of one production plan. The brief defines the sound and its evolution. The lyrics define the sections and the words. Start short, make the chorus structurally obvious, and revise identity before detail.

If you are deciding between Music 3's structured workflow and Suno's personalization features, continue with our MiniMax Music 3 vs Suno V5.5 comparison.