
A sung track: the prompt below set the arrangement and the script is what the model was given to sing. Copy the prompt, change a word and render your own.
What ACE-Step was given to sing, verbatim. Press play and it follows along.
Line positions are the rendered length divided by 3 lines. Neither model returns per-line timings and there is no forced aligner here, so the highlight follows the playhead approximately.
melodic drill, sliding 808 bass, skippy drill hi-hat pattern, dark piano loop, half-sung half-rapped male vocal, minor key, 142 BPM
'Half-sung half-rapped' describes melodic-drill delivery in four words. Put it in the prompt: ACE-Step reads that field as performance direction and the lyrics field as words to sing.
Listed as Apache 2.0 in the catalogue and flagged unverified there. Read the model card before shipping output commercially. Soundbox itself is MIT and imposes no terms of its own on what you generate.