music
Installation
SKILL.md
Music
Produce original music from a text prompt: full vocal songs with your lyrics, instrumental beds, short jingles, or background tracks. You supply a style/genre description (and lyrics when you want vocals), the model composes and renders the audio. A separate cover variant restyles an existing track instead of writing one from scratch.
Inputs to collect
- Style / genre. A short production-style description: genre, mood, tempo feel, instrumentation, era. This is required for every model.
- Vocals or instrumental. Whether the track should sing. If vocals, collect the lyrics (or ask the model to generate them from the prompt).
- Length / use. Full song, a 30-second bed, a short jingle. Affects model choice and
duration. - For a cover/restyle: the source audio (URL or UUID) and the target style. (Ask only if restyling.)
- Optional: BPM, musical key, vocal language, a seed for repeatability.
Models
- Full vocal song, promptable production: MiniMax Music 2.6 (
minimax:music@2.6) - natural-language or production-style prompts, reliable BPM/key following, lyrics or instrumental via settings. Strong default for finished songs. - Open, controllable songs + longer tracks: ACE-Step v1.5 Base (
runware:ace-step@v1.5-base) - vocals, explicit lyrics, BPM, key, vocal language, anddurationup to 5 minutes. Best when you need fine control or instrumental beds for video. - Fast iteration: ACE-Step v1.5 XL Turbo (
runware:ace-step@v1.5-xl-turbo) - accelerated 8-step inference for rapid drafts at higher capacity. Draft with this, finalize on Base or MiniMax. - Cover / restyle an existing track: MiniMax Music Cover (
minimax:music@cover) - keeps the original vocal melody, changes timbre, instrumentation, and genre from a text prompt. This isio:audio-to-audio, not text-to-music.