MiniMax Lyrics Generation API by MiniMax

MiniMax Lyrics Generation

Developer: MiniMax Model ID: minimax/lyrics-generation

Overview

MiniMax Lyrics Generation is a dedicated lyric-writing model that turns a one-line theme into a complete, professionally structured set of song lyrics. It does not just produce verse text: each result arrives with a song title, a set of style tags describing the intended genre and mood, and lyrics already segmented into named sections — [Intro], [Verse], [Pre Chorus], [Chorus], [Bridge], [Outro] and more.

That structure is the point. The section tags this model emits are the same tags MiniMax's music models read to control arrangement, so its output can be passed straight into minimax/music-3.0 as the lyrics input with no reformatting. The two models together form the standard two-step songwriting workflow: write the words, review or edit them, then compose the track.

Beyond writing from scratch, the model also works as a lyric editor — give it existing lyrics and an instruction, and it will rewrite, extend, or restructure them while keeping the parts you want intact.

How It Works

The model operates in one of two modes, selected by the mode parameter.

Write Full Song

In write_full_song mode, the prompt describes the song you want — theme, style, mood, scenario — and the model composes an entire lyric sheet from nothing. It decides the arrangement, writes each section, keeps the rhyme scheme and line lengths singable, and maintains a coherent narrative and emotional arc from opening to close. The lyrics field must be left empty in this mode.

If the prompt is left empty as well, the model generates a random song rather than failing — useful for ideation or sampling.

Edit Existing Lyrics

In edit mode, you supply existing text in the lyrics field and use the prompt as the editing instruction. This covers continuing an unfinished draft, rewriting a chorus, shifting the tone or register, tightening line lengths, translating the register to a different style, or adding missing sections. Because the model already understands section tags, it can restructure an arrangement — adding a bridge, splitting a verse — rather than only rewriting text in place.

Key Capabilities

• Structure-aware writing — output is segmented with section tags on their own lines, drawn from [Intro], [Verse], [Pre Chorus], [Chorus], [Post Chorus], [Hook], [Bridge], [Interlude], [Transition], [Build Up], [Break], [Inst], [Solo], and [Outro].
• Style tags returned alongside the lyrics — the model infers and returns descriptive tags (for example Pop, Summer Vibe, Romance, Lighthearted, Beach Pop), which double as a ready-made starting point for a music-generation prompt.
• Title generation, or title preservation — the model names the song for you, or keeps a title you supply unchanged.
• Two-way workflow — write from a theme, or edit and continue lyrics you already have, using the same endpoint.
• Singable output — lines are written with meter, phrasing, and rhyme in mind rather than as prose, so they fit a melody.
• Direct hand-off to music generation — the lyrics field of the result is format-compatible with the music models' lyrics input.

Input

Input Parameters

Parameter	Type	Default	Range / Options	Description
model	string	minimax/lyrics-generation	—	Required. Model name.
mode	string	write_full_song	write_full_song, edit	Required. Generation mode. write_full_song creates a complete new set of lyrics from the prompt; edit rewrites, continues or modifies the lyrics supplied in the lyrics field.
prompt	string	a lighthearted summer beach pop song about a seaside promise	≤ 2000 characters	Required. Song theme, style or editing instruction. If left empty, a random song is generated.
lyrics	string	""	≤ 3500 characters	Existing lyrics to continue or modify. Required when mode is edit, and must be left empty when mode is write_full_song, where the model writes the lyrics from scratch. Separate lines with a newline; supports section structure tags on their own line (e.g. [Verse], [Chorus]) to mark the arrangement of the song.
title	string	""	—	Song title. When provided, the generated result keeps this title unchanged.
enable_sync_mode	boolean	false	—	Wait for the result to be generated and uploaded before returning, so the lyrics are delivered directly in the response (API only).

Parameter Constraints

The schema ties lyrics to the selected mode:

• When mode is edit, lyrics is required and must contain at least one character.
• When mode is write_full_song, lyrics must be empty.

Output

The result is returned as a structured lyrics_result object with three fields:

Field	Type	Description
song_title	string	Generated song title. If title was provided in the request, it is preserved — for example "Summer Breeze Promise".
style_tags	array of strings	Style tags describing the song, for example ["Pop", "Summer Vibe", "Romance", "Lighthearted", "Beach Pop"].
lyrics	string	The generated lyrics, with lines separated by newlines and sections marked by structure tags such as [Intro], [Verse], [Pre Chorus], [Chorus], [Bridge] and [Outro]. Directly compatible with the lyrics field of the music generation models.

Requests are asynchronous by default: the response carries a request ID that can be polled until the lyrics are ready, and enabling sync mode returns the finished result directly. The response also reports task status, executionTime, and an inference timing breakdown.

Use Cases

• Step one of a two-step songwriting pipeline — generate lyrics, let a human review or adjust them, then compose the track with a music model.
• Lyric ideation — produce several structurally complete drafts on a theme and pick the strongest, instead of starting from a blank page.
• Rewriting and polishing — tighten a chorus, fix an awkward verse, or shift a draft's tone without rewriting the whole song.
• Extending unfinished drafts — continue a fragment into a full arrangement with the missing sections added.
• Personalized and occasion songs — birthday, wedding, or dedication lyrics generated from a short brief.
• Localized and stylistic variants — regenerate the same theme in a different genre register or voice for A/B testing.

Pricing

Pricing for Lyrics Generation is a flat rate per song. It does not vary with mode, prompt length, the length of the lyrics submitted for editing, or the length of the lyrics produced — writing a full song from scratch and editing an existing one are billed identically.

SKUs

SKU	Description	Unit Price
sku_base	Per song	$0.01

Formula

cost = sku_base

The formula is the single base SKU, so each successful request costs $0.01, and total spend is the number of requests multiplied by that rate.

Examples

Scenario	Calculation	Cost
Write a full song from a theme	0.01	$0.0100
Edit or continue existing lyrics	0.01	$0.0100
Random song (empty prompt)	0.01	$0.0100
10 lyric drafts for comparison	0.01 × 10	$0.1000
100 requests	0.01 × 100	$1.0000

At a cent per request, generating and comparing multiple lyric drafts before committing to a music generation is inexpensive relative to the composition step itself, which makes an iterate-on-lyrics-first workflow the economical default.

Related Models

minimax/music-3.0 is the composition counterpart to this model: it takes the lyrics string produced here — section tags included — and returns a complete, fully produced song of up to five minutes. The style_tags returned by this model make a good seed for that model's prompt. Use this model when you want visibility into and control over the words before any audio is generated.