How to Use MiniMax Music 2.5 in AI FILMS Studio: Lyrics to Music Guide

MiniMax Music 2.5 writes a complete song from a style description and a set of lyrics. It returns one mixed track with vocals and instrumentation together, rather than stems you assemble yourself. The model suits filmmakers who need an original song for a scene, a title sequence, or a trailer bed without clearing rights on a commercial recording.

This guide covers the model as it appears in the AI FILMS Studio music workspace and in the Nodes Graph Editor.

Where to Find It

Open the AI FILMS Studio music workspace or click Create Music in the top navigation.

One point causes confusion. MiniMax Music 2.5 sits under Lyric to Music, not under Text to Music. The Text to Music generation type offers a different model. If you open the model picker and cannot find MiniMax Music 2.5, check the Select Generation Type dropdown first.

Selecting the Model

With Lyric to Music set, open Select Model. The picker is searchable and lists two options.

Writing the Style Prompt

The Style Prompt field defines genre, mood and instrumentation. It accepts up to 3,000 characters, though short comma separated lists work better than paragraphs.

A style prompt that names instruments and production character outperforms one that names only a genre. "rock" produces something generic. "sub bass 808s, driving mid tempo live drums, sweeping symphonic strings, dark melancholic atmospheric pads" tells the model what to build.

Writing the Lyrics

The Lyrics field is where MiniMax Music 2.5 differs most from a text to music model. It expects real lyrics, and the field is required. The counter caps at 3,000 characters.

The ## markers work as a pair. Wrap a passage in ## on both sides and the model treats the text inside as an instrumental direction rather than words to sing. A working example looks like this:

##lyrics## [Instrumental Intro: Heavy analog synthesizer arpeggio,
driving 16th-note bassline, industrial acoustic drum entry] ##lyrics##

[verse]
The sun rises on the broken ground
Silence screams without a sound

[chorus]
I stand alone against the gale
Through the storm, I will prevail

Structure markers change the arrangement, not just the vocal delivery. A lyric sheet with a marked chorus produces a track that returns to a recognisable hook. The same words with no markers produce something flatter.

The Three Audio Output Controls

MiniMax Music 2.5 exposes three output settings that ACE-Step 1.5 does not. All three affect the file you download, not the composition.

Parameter Reference

Control	Values	Default	What it changes
Select Generation Type	Text to Music, Lyric to Music	Text to Music	Which models are available. MiniMax Music 2.5 needs Lyric to Music
Select Model	ACE-Step 1.5, MiniMax Music 2.5	ACE-Step 1.5	Which controls appear below
Style Prompt	Up to 3,000 characters	Empty	Genre, mood, instrumentation
Lyrics	Up to 3,000 characters, required	Empty	The words sung, plus structure and arrangement
Select Sample Rate	16000, 24000, 32000, 44100 Hz	44100 Hz	Frequency detail in the output file
Select Bitrate	32, 60, 64, 128, 256 kbps	256 kbps	Compression and file size
Select Audio Format	MP3, PCM, FLAC	MP3	Container and compression type

Each song generation costs 150 credits, and the figure appears in the form before you submit.

Running It in the Nodes Graph Editor

The Nodes Graph Editor exposes the same model through a node named Text To Music. The node name refers to the node family, and the model selector inside it still offers MiniMax Music 2.5.

The node route is worth using when the song is one step in a longer chain. Generating a track and passing it straight into a lipsync or video node keeps the whole sequence in one graph.

Limitations

The lyrics field is required. MiniMax Music 2.5 will not produce an instrumental from an empty lyrics field. For an instrumental, either wrap the whole lyric in ## accompaniment markers or switch to a model that treats lyrics as optional.

There is no duration control. Track length follows from the lyrics and structure you supply. Longer lyric sheets with more marked sections produce longer tracks. If you need an exact runtime, generate and then trim in the editor.

There is no seed field. Two runs with identical settings will not return identical audio. Download anything you want to keep before running again.

Output is a single mixed file. The model does not return separated stems, so you cannot rebalance vocals against instrumentation afterwards. Plan the mix through the style prompt instead.

Tips for Better Results

Name instruments, not just genres. The style prompt responds to specifics. Naming the drum character, the bass register and the pad texture changes the output far more than adding a third genre tag.

Mark the structure every time. Even a short lyric benefits from [verse] and [chorus] markers. They tell the model where the arrangement should lift and where it should return.

Keep 44100 Hz and 256 kbps for anything you will edit. Lower settings are for previewing ideas quickly. The quality difference is audible once a track sits under dialogue.

Write the lyric first, then the style. Deciding the song structure before describing the production tends to produce a track that holds together, because the style prompt can then describe how that specific arrangement should sound.

For a model on the same generation type that accepts optional lyrics and gives you a duration slider and a seed, ACE-Step 1.5 runs open source under an MIT license and scores above Suno v5 on SongEval benchmarks. For style driven generation without writing lyrics at all, the Suno text to music workflow in AI FILMS Studio covers the Text to Music generation type. Filmmakers who also need sound design can compare Stable Audio 3 for open weight music and sound effects.

---

Sources

MiniMax | AI FILMS Studio | SongEval