Claude music cover

Skill AgriciDaniel/claude-music/skills/claude-music-cover

Creates cover versions and style transfers of existing songs using ACE-Step 1.5. Takes a reference audio file and generates a new version with different style, genre, or vocal characteristics while preserving musical structure.From its SKILL.md

Install

npx -y skills add AgriciDaniel/claude-music --skill claude-music-cover

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

4.3 KB, as published. Nobody here has run it

claude-music-cover — Cover/Style Transfer

Worked example (input → command → output)

User: "Turn this pop song into a jazz cover" (hands over ~/Music/my_song.flac)

Command:

bash ~/.claude/skills/claude-music/scripts/music_engine.sh \
  --quality standard \
  cover \
  --src-audio "$HOME/Music/my_song.flac" \
  --caption "smooth jazz, saxophone, piano trio, warm, live recording feel" \
  --cover-strength 0.5

Expected JSON on stdout:

{
  "success": true,
  "task_type": "cover",
  "model": "acestep-v15-turbo",
  "outputs": [
    {"path": "/home/$USER/Music/claude-music-output/cover_20260416_121040_01.flac",
     "seed": 4921, "size_mb": 2.47, "index": 1},
    {"path": "/home/$USER/Music/claude-music-output/cover_20260416_121040_02.flac",
     "seed": 81244, "size_mb": 2.52, "index": 2}
  ],
  "params": {
    "src_audio": "/home/$USER/Music/my_song.flac",
    "cover_strength": 0.5,
    "caption": "smooth jazz, saxophone, piano trio, warm, live recording feel"
  },
  "timing": {"generation_sec": 29.4},
  "count": 2
}

Note: --cover-strength maps to ACE-Step's cover_noise_strength (0.0 = fully reimagine source, 1.0 = stay close to source). The code passes src_audio + cover_noise_strength — the older reference_audio/audio_cover_strength field names are wrong and guarded by a regression test.

Pre-Flight

Verify source audio exists: bash ~/.claude/skills/claude-music/scripts/preflight.sh "$SRC_AUDIO" "$OUTPUT"
Check GPU: bash ~/.claude/skills/claude-music/scripts/detect_gpu.sh

Usage

# Jazz cover of a pop song
bash ~/.claude/skills/claude-music/scripts/music_engine.sh cover \
  --src-audio /path/to/original.mp3 \
  --caption "smooth jazz, saxophone, piano trio" \
  --cover-strength 0.5

# Same genre, different mood
bash ~/.claude/skills/claude-music/scripts/music_engine.sh cover \
  --src-audio song.flac \
  --caption "acoustic version, stripped back, intimate" \
  --cover-strength 0.7

# Complete reimagination
bash ~/.claude/skills/claude-music/scripts/music_engine.sh cover \
  --src-audio track.wav \
  --caption "electronic, synthwave, 80s retro" \
  --cover-strength 0.2

# Cover with new lyrics
bash ~/.claude/skills/claude-music/scripts/music_engine.sh cover \
  --src-audio original.mp3 \
  --caption "rock, electric guitar, powerful male vocal" \
  --lyrics "[Verse] New lyrics here..." \
  --cover-strength 0.5

Cover Strength Guide

--cover-strength controls cover_noise_strength (how much of the source audio's latent is preserved):

Value	Effect	Use When
0.0	Pure noise — complete reimagination, only caption guides output	Major genre transformation
0.2-0.3	Loose inspiration — new arrangement, some structural similarity	Different genre/mood
0.4-0.5	Moderate transfer — recognizable structure, new details	Balanced cover (default)
0.6-0.7	Faithful adaptation — preserves melody/rhythm, changes timbre	Same genre, different voice
0.8-1.0	Close to source — minimal changes, mostly regeneration	Voice swap, subtle tweaks

Parameters

Parameter	Flag	Default	Description
Source audio	`--src-audio`	required	Path to reference audio file
Caption	`--caption` / `-c`	`""`	New style description
Lyrics	`--lyrics` / `-l`	`""`	New lyrics (optional)
Cover strength	`--cover-strength`	0.5	Noise preservation (0.0=pure noise/reimagine, 1.0=closest to source)
Quality	`--quality`	standard	draft/standard/high/max
Format	`--format`	flac	Output audio format

Reference Audio Tips

Supported formats: MP3, WAV, FLAC, OGG, AAC, M4A
Optimal length: Match your target duration (model uses the full reference)
Quality matters: Higher quality reference = better style extraction
Loudness: Normalize reference to ~-14 LUFS for consistency