Worked example (input → command → output)
User: "surprise me"
Claude picks: Lo-fi (from the pool below), composes a caption.
Command:
bash ~/.claude/skills/claude-music/scripts/music_engine.sh \
--quality draft \
generate \
--caption "lo-fi hip-hop, chill, atmospheric, vinyl crackle, mellow piano" \
--instrumental --duration 60
Expected JSON on stdout (batch=4 from draft preset):
{
"success": true,
"task_type": "text2music",
"model": "acestep-v15-turbo",
"outputs": [
{"path": ".../text2music_20260416_121530_01.flac", "seed": 81, "index": 1},
{"path": ".../text2music_20260416_121530_02.flac", "seed": 84012, "index": 2},
{"path": ".../text2music_20260416_121530_03.flac", "seed": 44190, "index": 3},
{"path": ".../text2music_20260416_121530_04.flac", "seed": 2031, "index": 4}
],
"timing": {"generation_sec": 14.8},
"count": 4
}
Then present all four paths and ask which to iterate on.
Workflow
- Pick a random genre from the list below
- Craft a caption using genre-appropriate tags
- Generate with draft quality (4 variants, ~15s)
Genre Pool
Pick randomly from: Pop, Rock, Jazz, Lo-fi, Electronic, Hip-hop, R&B, Ambient, Folk, Synthwave, Bossa Nova, Afrobeat, Funk, Soul, Blues, Country, Reggae, Classical Piano, Cinematic, Chillwave, House, DnB, Trap, Phonk
Quick Generate
# Instrumental random (fastest)
bash ~/.claude/skills/claude-music/scripts/music_engine.sh generate \
--caption "<genre tags here>" \
--instrumental --quality draft --duration 60
# With vocals (Claude writes a quick verse + chorus)
bash ~/.claude/skills/claude-music/scripts/music_engine.sh generate \
--caption "<genre tags>" \
--lyrics "[Verse] <quick lyrics> [Chorus] <catchy hook>" \
--quality draft --duration 60
After Generation
Present all 4 variants with paths and suggest: "Want me to refine any of these? I can repaint sections, change the style, or generate more in this direction."