Write the words, then hear them sung

The generation type dropdown, with Lyric to Music selected.

The music workspace opens on a dropdown holding Text to Music and Lyric to Music. Each carries its own model roster, so the choice here decides which models the picker below will show you.

The name understates what this generation type does. Lyrics are optional on one of the two models here, so this is also where you come for an instrumental with a duration you control.

ACE-Step 1.5 carries a Default and Fast label and is already selected. MiniMax Music 2.5 carries High Quality and Latest. The picker is searchable, which matters less with two entries than it will later.

This choice rewrites the bottom half of the form. ACE-Step 1.5 gives you a duration slider and a seed. MiniMax Music 2.5 replaces both with sample rate, bitrate and audio format. There is no overlap between the two sets.

Two models, and the labels beside them describe the trade rather than the ranking.

The style field, whose label changes with the selected model.

The label reads Style Tags on ACE-Step 1.5 and Style Prompt on MiniMax Music 2.5. That is not cosmetic. The first wants comma separated tags, the second accepts a fuller description, and both hold 3,000 characters.

The buttons below append without typing, covering rap, k-pop, hard house, synth-pop, classical, jazz, country, alternate-rock, european, rock, R&B, EDM, reggae, blues, folk, metal, punk, disco, soul and funk. Adding a tempo in beats per minute alongside them gives the model a target that a genre tag cannot imply.

On ACE-Step 1.5 this field is optional and the label says so. Leaving it empty returns an instrumental with no vocal line at all, which is the single most useful behaviour on this page for anyone scoring picture.

Eight buttons sit underneath. Intro, Verse, Chorus, Bridge and Outro insert a lowercase structure marker in square brackets. New Line inserts a single break, Pause inserts a double break that reads as a rest, and Accompaniment inserts a ## marker used in pairs to wrap an instrumental direction.

The lyrics field with its eight markers. Note the optional label on ACE-Step 1.5.

Duration, on ACE-Step 1.5 only. Marks at 1s, 60s, 120s and 240s.

The slider runs from 1 second to 240 seconds and starts at 60. Four minutes covers a full song, and the low end is short enough to generate a sting or a transition element.

Set this to match the cut rather than generating long and trimming. The model paces an arrangement across whatever length it is given, so a 30 second setting resolves at 30 seconds instead of producing a four minute piece cut short.

Seed defaults to -1, which the helper text under the field confirms means random. The dice button on the right fills in a number for you.

A fixed seed is what makes iteration possible. Hold the seed and change one style tag, and you hear what that tag did. With a random seed every run is a different piece and comparison tells you nothing.

Seed, on ACE-Step 1.5 only. The dice button fills in a random value.

Sample Rate, on MiniMax Music 2.5 only. 44100 Hz is the default.

Switching to MiniMax Music 2.5 removes the duration slider and the seed field, and puts three output controls in their place. The first is Sample Rate, offering 16000 Hz, 24000 Hz, 32000 Hz and 44100 Hz.

44100 Hz is the default and the standard for streaming and broadcast delivery, so it drops into a timeline without conversion. The lower rates exist for speech style output and small file delivery.

Bitrate offers 32, 60, 64, 128 and 256 kbps, and defaults to the top of that range. It controls compression, so it decides both the file size and how much detail survives in the mix.

Stay at 256 kbps for anything heading into an edit. The lower settings are for previewing an idea quickly, and the difference becomes audible once a track sits under dialogue.

Bitrate, on MiniMax Music 2.5 only. 256 kbps is the default.

Audio Format, on MiniMax Music 2.5 only. Three containers.

The last of the three controls offers MP3, PCM and FLAC, with MP3 as the default. This decides the container rather than anything about the music itself.

MP3 is compressed and opens anywhere. FLAC is lossless compression, so it keeps full quality at roughly half the size of raw audio. PCM is uncompressed and produces the largest files.

The generated track appears on the right with a player, a download button and a delete button. Each run returns a single file rather than a pair, which is where this generation type differs from Text to Music.

Track length on MiniMax Music 2.5 follows from the lyrics and structure you supplied, since it has no duration control. The example here runs 2 minutes 48 seconds from a lyric sheet with marked sections.

The finished result. One file, with a player, a download and a delete.