build-music

Describe the genre, mood, tempo, and length, and the skill composes chords, melody, bass, and drums with genre-aware helpers, then renders a stereo WAV into Assets/GeneratedSFX. The output folder is shared with build-sfx, so the Lens picks it up automatically.

Supported genre tags include pop, sad-pop, lofi, lofi-jazz, jazz, cinematic-epic, cinematic-melancholy, dreamy, folk, edm-uplift, dark, rnb, synthwave, funk, and ambient. Unknown tags fall back to pop, and each genre drives its own progression, key, voicing, tempo range, and drum groove pools.

No. All audio is produced by pure DSP synthesis (additive, FM, waveguide, Karplus-Strong voices) running locally in Node.js. The output is sample-free and license-clean, and the entire pipeline works fully offline.

The renderer only plays numeric-MIDI events, so string drum names like 'bd' or 'sn' in a hand-written pattern produce silence. Use composeDrums, which emits numeric-MIDI track descriptors, or map names through DRUM_MAP before rendering.

Yes. The compose helpers log a meta object containing the seed, key, progression, voicing, and groove picks. Pass that seed back into composeChords, composeDrums, or composeBass to pin the take; omitting seeds gives a fresh variation every run.

Pure DSP synthesis cannot convincingly fake certain instruments, such as bowed string sections, because it lacks the physical modeling and convolution those timbres require. For realistic strings or specific recorded sounds, use sampled assets instead of synthesis.