Agent skill · recoupable
recoup-minimax-music-3
Generate an original song with MiniMax Music 3 on fal, for a film, an ad, a demo or a release. Covers the two inputs the model takes (a structured Caption and tagged Lyrics), how to write lyrics that survive generation, the exact fal call and its cost, and the checks to run on the returned audio before anything is built on top of it. Use when asked to "make a song", "generate a track", "write the song for this video", "score this", or when a music video needs a song that does not exist yet. For an artist's already-released song use recoup-internal-marketing's music-video.md instead; for analysing an existing recording use recoup-song-analyze-audio.
What it needs
About 4k tokens when loaded.
What this skill does
MiniMax Music 3 Two inputs drive this model and they do different jobs: Input fal field Owns --- --- --- Caption prompt the music: genre, tempo, key, vocals, arrangement Lyrics lyrics the words and the song's structure, via bracketed tags Naming trap. The open-weights repo and the model card call these instructions and input. On fal they are prompt and lyrics. Same two things, different names. Do not write the Caption by hand. Delegate it. MiniMax ships an official skill for exactly this, with a thousand style templates and a genre router: Hand it your brief and it returns a caption under three required headings: Global Metadata, Vocal Details, Arrangement. That is the format the model was trained to read, so use it. It targets roughly 250 to 450 words. It will not write your lyrics. It reads bracketed tags only and is forbidden from quoting or paraphrasing lyric content. Lyrics are this skill's job, below. If you are writing a caption without it, the minimum viable shape is a labelled header, then instrumentation, then per-section arrangement: That is the house precedent (content/rough-draft, 2026-08-25) and it worked. Label BPM and key explicitly; burying tempo in prose is weaker. Lyrics: the tags are the structure Only these tags are recognised. Anything else is sung aloud. [Intro] [Verse] [Pre-Chorus] [Chorus] [Post-Chorus] [Bridge] [Instrumental] [Solo] [Outro] Put each tag on its own line, lowercase or titlecase both work. For an instrumental, omit words entirely and carry the piece on [Instrumental] and [Solo]. Write to the line budget or the model will improvise About ten sung lines per sixty seconds. Under-supply lyrics and the model does not stop early, it fills the remaining time by repeating and mangling your last section. Measured 2026-09-02: a 26-line lyric at duration: 110 sang cleanly to 89.7s, then looped a garbled version of the final line three times out to 106.8s. …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills recoupable/recoup-minimax-music-3