Claudewave

claudewave

A [Claude Code](https://claude.com/claude-code) skill for building vaporwave/slushwave/citypop/chillwave remixes and sound-reactive ASCII music videos of existing songs.

Claude handles the whole pipeline: stem separation, per-instrument MIDI transcription, beat-tracked arrangement, programmatic synth rebuild, genre-specific DSP, and 1080p60 ASCII visualizers with per-song reflections.

What it does

Given an mp3/wav, Claude will:

1. Separate it into 4 stems (demucs).
2. Transcribe bass + harmonic instruments to MIDI (basic-pitch) for pitch/chord info.
3. Beat-track the original for the rhythm grid.
4. Detect chord progression (including jazz extensions).
5. (Optional) Generate AI atmosphere pads + m2m vocal style transfer via a local ComfyUI server with ACE-Step.
6. (Optional) Extract monophonic melodies (whistle/lead) with pYIN.
7. Rebuild the track with programmatic synth voices (FM Rhodes, slap bass, sub, bell, Juno pad, whistle, vocoder).
8. Apply genre-specific DSP (chorus/phaser/reverb/tape wobble/sidechain pump/chopped hook loops).
9. Render a 1080p60 ASCII music video with scene visuals + a reflection panel with AI/lyric-related commentary.
10. Mux the final video with MP3-in-MP4 for broad compatibility.

Install

See [`setup.md`](setup.md) for full install commands. Summary:

• ffmpeg (with libx264, libmp3lame, optionally h264_nvenc for GPU encode)
• Python 3.13 main env: librosa soundfile scipy numpy pedalboard demucs Pillow music21
• Python 3.12 side env (for basic-pitch which has deps that don't build on 3.13): basic-pitch onnxruntime pretty_midi librosa soundfile scipy
• Optional: a local ComfyUI server at 127.0.0.1:8188 with ace_step_1.5_turbo_aio.safetensors and ace_step_v1_3.5b.safetensors for AI music generation

Use

Add the skill directory to Claude Code's skills path, then just ask:

make a slushwave remix of this track with a music video

turn this into a citypop cover

vaporwave t