MiniMax Music 3Complete Songs From Lyrics And A Description
MiniMax's flagship music model writes, performs, and produces in a single pass. Give it lyrics with section tags and a description of the sound you want, and it returns a complete track with lead vocals, harmonies, and a full instrumental arrangement, up to five minutes long.
What makes MiniMax Music 3 different
Complete songs, up to five minutes
Music 3 generates a finished track in one pass rather than a loopable clip you have to arrange yourself.
- Set any length up to 300 seconds with the duration parameter.
- Intro, verses, choruses, bridge, and outro arrive already sequenced.
- The model can resolve a song early; the true length comes back in the response.
- Output is a 44.1 kHz 16-bit stereo WAV, ready to master.
You write the words, the model sings them
Pass real lyrics and Music 3 performs them with an intelligible lead vocal instead of the wordless vocal texture most music models produce.
- Section tags such as [verse], [chorus], [bridge], and [outro] control the arc of the song.
- Each tag sits on its own line, and the model builds the arrangement around it.
- Use [instrumental] or [solo] to hand a section over to the band.
- Lead vocals, stacked harmonies, and ad-libs are generated together.
Describe the record you want made
The prompt works as a production brief, and genre is the lever that moves it furthest.
- Lead with the genre. Style labels steer the take harder than anything else in the prompt.
- Set BPM and key to fix the musical foundation.
- Say how the vocals should stack: a solo lead, voices trading lines, or full group harmony.
- Name what you do not want. Ruling instruments out explicitly is what unlocks sparse and a cappella arrangements.
Re-roll a take or lock one in
Every generation returns the seed it used, so a track you like is repeatable and a track you nearly like is tunable.
- Reuse a returned seed to regenerate the same take.
- Hold the seed and edit the lyrics to keep the production while changing the words.
- Raise num_inference_steps to trade speed for fidelity, up to 100 per chunk.
- Adjust guidance_scale to tighten or loosen how literally the prompt is followed.
One model, every genre
Ten songs, ten genres, one model. Every track below came out of Music 3 from the prompt on its card, and each performer is animated from that same audio, so the singing you see matches the singing you hear. Copy any prompt and run it yourself in the MiniMax Music 3 playground.
Neon Hours
"Genre: glossy synth-pop. BPM: 112. Key: F# minor. Bright, euphoric, wide-open chorus with a nostalgic undertow. Vocals: female lead, clear and forward, doubled an octave up on the hook with airy stacked harmonies. Arrangement: pulsing analog synth bass, gated reverb drums, shimmering pad chords, a single plucked arpeggio hook that carries the chorus."
Corner Store Prophet
"Genre: boom-bap hip hop with a soulful R&B chorus. BPM: 88. Key: D minor. Confident, gritty, head-nod groove. Vocals: male rapper, mid-range, punchy and conversational, tight ad-libs behind the bars, opening out into a smooth sung hook. Arrangement: thick swung drums, warm rolling bass, mellow keys underneath."
Two-Lane Sunset
"Genre: modern country rock with an outlaw edge. BPM: 96. Key: G major. Warm, road-worn, big singalong chorus. Vocals: male lead with a natural rasp and a slight drawl, tight family-style harmonies on the chorus. Arrangement: strummed acoustic under twangy telecaster licks, warm bass and a full drum backbeat."
Louder Than The Dark
"Genre: 1980s arena hard rock, in the style of Guns N' Roses, Van Halen and Bon Jovi. BPM: 132. Key: E major. Swaggering, anthemic, stadium-sized. Vocals: male lead only, gritty raspy high tenor with a bluesy wail, clearly enunciated words, big gang-vocal chorus behind him. Arrangement: crunchy overdriven Les Paul riffs, a screaming twin-guitar solo, live drums with big reverb, Hammond organ pad under the chorus."
Lightstick Ocean
"Genre: Korean K-pop idol dance track, sung in Korean with English hooks. BPM: 124. Key: B minor. Sleek, punchy, polished Seoul idol production. Vocals: young Korean female idol lead singing in Korean, bright and clear, group chant on the hook. Arrangement: hard trap-influenced drums, distorted synth bass, glassy plucks, brass stabs on the drop."
Bedroom Ceiling
"Genre: mellow indie folk-pop. BPM: 96. Key: D major. Warm, intimate, easy. Vocals: male lead singing smoothly and cleanly in a soft mid register, velvety and gentle, no rasp or huskiness at all. Arrangement: softly fingerpicked nylon and warm electric piano, muted low-mid tones only, no bright jangly guitar, no piercing treble, no aggressive strumming. Soft muted drums, warm upright bass."
Paper Boats
"Genre: intimate acoustic folk. BPM: 72. Key: C major. Fragile, close, and unhurried, barely rising by the end. Vocals: soft female lead, breathy and very close to the microphone, audible breaths, one quiet harmony line entering late. Arrangement: fingerpicked nylon-string guitar, sparse upright piano notes, faint room tone, no drums."
Say It Back
"Genre: 1990s boy band pop, in the style of a five-piece vocal harmony group. Multiple male voices singing together in close four and five part harmony on every chorus, layered and stacked, trading solo lines in the verse. Smooth tenor lead with a low baritone underneath and a falsetto on top. Earnest romantic pop ballad production."
Iron Winter
"Genre: aggressive thrash metal. BPM: 160. Key: D minor. Ferocious, cold, and relentless. Vocals: harsh male screams in a piercing high register, layered. Arrangement: down-tuned tremolo-picked guitars, double-kick drum blast sections, twin harmonized guitar lead, deep distorted bass."
Kyrie For The Morning
"A cappella choral music. Unaccompanied voices only. There are NO instruments of any kind: no drums, no bass, no guitar, no synth, no piano, no organ. Silence except for human voices. A large mixed choir of many singers performs in four-part harmony, soprano alto tenor bass, in a stone cathedral with long natural reverb. Slow, sustained, reverent."
MiniMax music models on fal
Music 3 is the newest generation of MiniMax's music models on fal. Every version is available through the same serverless API.
MiniMax Music 3 is a high-performance music generation model for creating complete songs up to five minutes long
MiniMax Music 2.6 creates complete tracks with singing, backing music, and detailed arrangements from lyrics and a style description.
MiniMax Music 2.5 creates complete tracks with singing, backing music, and detailed arrangements from lyrics and a style description.
How to get access to the MiniMax Music 3 API
One serverless endpoint takes a prompt and lyrics and returns a finished song. The client manages the submit protocol, streaming status updates and returning the audio when generation completes.
import { fal } from "@fal-ai/client";
const result = await fal.subscribe("minimax/music-3", {
input: {
prompt:
"Genre: arena rock anthem. BPM: 128. Key: E major. Soaring and " +
"triumphant with a huge festival chorus. Vocals: powerful male lead, " +
"belted, wide stacked gang harmonies on the hook. Arrangement: driving " +
"distorted guitars, punchy live drums, bright synth pad, thick low end.",
lyrics: [
"[verse]",
"We were static on a broken screen",
"Counting hours in a town asleep",
"[chorus]",
"Turn it up until the silence breaks",
"We are louder than the dark we make",
].join("\n"),
duration: 90,
},
logs: true,
onQueueUpdate: (update) => {
if (update.status === "IN_PROGRESS") {
update.logs.map((log) => log.message).forEach(console.log);
}
},
});
// 44.1 kHz 16-bit stereo WAV, plus the seed and the true duration.
console.log(result.data.audio.url);
console.log(result.data.seed, result.data.duration);
Common questions about MiniMax Music 3
What is MiniMax Music 3?
MiniMax Music 3 is a text-to-music model that generates a complete song from written lyrics and a description of the sound. It produces lead vocals, backing harmonies, and a full instrumental arrangement together as one finished track of up to five minutes, returned as a 44.1 kHz 16-bit stereo WAV. It is available through a serverless API on fal.ai.
How long can a MiniMax Music 3 song be?
Up to 300 seconds, or five minutes. The duration parameter sets an upper bound rather than an exact length: the model may resolve a song earlier if the arrangement finishes, and the actual length is returned in the response alongside the audio. The default is 60 seconds.
How do I control the structure of the song?
Write section tags into the lyrics, each on its own line. Music 3 recognizes [intro], [verse], [pre-chorus], [chorus], [post-chorus], [bridge], [instrumental], [solo], and [outro], and arranges the track around them. Text placed on the same line as a tag is dropped by the model's input contract, so always start the lyric on the line below the tag.
What makes a good prompt?
Treat the prompt as a production brief rather than a mood word. Name the genre, BPM, and key; describe the vocalist's range, texture, and delivery; then specify the arrangement instrument by instrument and when each one enters. For tighter control, use a structured caption that carries global metadata, vocal details, and a section-by-section plan in a single prompt.
Can I generate the same song twice?
Yes. Every response includes the seed used for that generation, and passing it back reproduces the take. This also makes targeted edits possible: hold the seed and change the lyrics to keep the production while swapping the words, or hold both and raise num_inference_steps to render the same arrangement at higher fidelity.
Can Music 3 generate instrumental tracks?
Yes. Describe the instrumental you want in the prompt and use [instrumental] and [solo] section tags to hand the arrangement to the band. The lyrics field is required by the API, so an instrumental is expressed as a lyric body made up of instrumental section tags rather than an empty field.
What audio format does it return?
A 44.1 kHz 16-bit stereo WAV file, delivered as a URL in the response along with the seed and the true duration of the track. Because it is uncompressed CD-quality stereo, it goes straight into a DAW or a mastering chain without a lossy intermediate step.
Can I use Music 3 to score a video?
Yes, and it is a common pairing on fal. Generate the footage with a video model such as Seedance 2.0 or MiniMax H3, then generate a track with Music 3 and lay it underneath. Set the duration to match your edit, and use the returned seed to re-render the same arrangement if the cut changes. The example clips on this page were built exactly that way.
How does Music 3 differ from the other audio models on fal?
Music 3 is built for songs: lyrics in, a finished record with vocals and a full arrangement out. Seed Audio 1.0 is built for scenes, generating dialogue, sound effects, and music together, and ElevenLabs covers speech and voice work. Reach for Music 3 when the deliverable is a song rather than a soundscape or a voiceover.
How do I get started with the MiniMax Music 3 API?
Install the fal.ai SDK (Python or JavaScript), grab an API key from your dashboard, and call the minimax/music-3 endpoint with a prompt and lyrics. The API is serverless, so there is no infrastructure to set up. Check the API documentation for every available parameter.
Can I use MiniMax Music 3 for commercial projects?
Yes. Content generated through the fal.ai API can be used in commercial projects. Check fal.ai's terms of service for full details on usage rights and licensing.
Get in touch about MiniMax Music 3
Want to learn more about integrating MiniMax Music 3 into your workflow? Leave your details and our team will reach out.