Emocute Visual — Turn Claude Code into a music video engineer

A toolkit you drive through Claude Code to make music videos.

Just saying "make an MV for this track" works. Or you can be specific in plain English, like:

"It's ambient, so use moonlit_sea_night. I also want to feature the chord progression, so try cof_pianoroll_green too and let me compare."

Claude handles audio analysis, template selection, build, and export end-to-end and returns a draft. From there you refine through conversation — which template, what to emphasize, what to swap. That's the harness.

Works through conversation

• "Use moonlit_sea_night for this ambient track"
• "I want the chord progression to be the star. Build with cof_pianoroll_green too so I can compare"
• "Time-stretch to 0.95×, preserve the original key"
• "Generate Spotify Canvases (9:16 / 8s) for every track on the album in one batch"

The toolkit ships with 39 commands, 5 templates, and 9 parts, and Claude composes them to fulfill your request. You don't need to learn the CLI (though you can).

Driving it yourself, through Claude

Even when you want to direct your own staging — not hand it off completely — Claude is still your starting point:

• "Detect chord progression, BPM, and key for this track" → Claude runs detection and returns results
• "Emphasize this chord transition on the circle of fifths" → Claude rewrites parameters for that span
• "Push the colors cooler, calm down the wave motion" → Claude rewrites the matching template and rebuilds
• "I want to feed the chord progression I wrote in my DAW to a template" → MIDI import (midi-chords command)

You're not stuck with existing template combinations. Tell Claude "make the particles coarser," "explode the waveform only on the chorus," or "draw it as this shape instead of the circle of fifths" — Claude rewrites the shaders and helps you develop your own visual language. Source code is included; let Claude decide what to change and where.

Extracted music features (chord progression, BPM, key, audio descriptors) are stored as JSON, so you're free to take them into a DAW or another visualizer.

Three export pipelines

• autopilot — One MP3 in, a finished MV out (analysis → build → MP4)
• build --auto-record — Single click in the browser, auto-downloads MP4 when the song ends
• webm-to-mp4 — Re-encodes WebM to MP4

Source-included, not a black box.

Core features

• Circle-of-fifths chord visuals — Templates and parts render chord progressions as transitions on the circle of fifths (templates: cof_pad_keyboard, cof_pianoroll_green; parts: cof_chord_key, cof_bass_wave). Music theory becomes the visual language.
• Hardware-UI tribute parts — Stratum Wave (layered spectral terrain), MPC Pads (4×4 pad matrix), Step Sequencer (16-step × 4-track), Lissajous (XY phase scope). Visual vocabulary that reads naturally to gear-literate listeners.
• Chord recognition engine — Built on BTC (Bi-directional Transformer for Chord recognition) with a 402-chord dictionary (maj9, 7b9, m11, 13#11, etc.), inversion / missing-tone / extra-tone score handling via ChordMatch, and sub-bin pitch detection.
• Spotify Canvas batch mode — Generate Canvases (9:16, 8s loops) for an entire album in one batch. Per Spotify's own data, Canvas drives +5% streams, +145% shares, and +20% playlist adds.
• MIDI → chord progression (midi-chords) — Generate chords.json from MIDI exported by your DAW. More reliable than auto-detection, and lets you feed hand-written chord progressions straight into a template.
• Helper commands — LRC / SRT lyric import, crossfade joining.

What's in the bundle

• mvtool.py + commands/ (39 commands / ~14,000 lines of Python)
• harmony.py + chord_defs.py (402-chord dictionary)
• detect.py (chord recognition engine)
• Curated template library (WebGL2 shaders, JS-error-checked)
• Sample track + MIDI (sample/Emocute - Études Impossible.mp3 + bundled MIDI ZIP) — a felt-piano IDM instrumental so you can do an end-to-end dry run the moment you unzip.
• 22 documentation files + CLAUDE.md (optimized for Claude Code integration)
• License (commercial use OK / redistribution prohibited)

How to use it

1. Unzip
2. Launch Claude Code (claude command) in the unzipped folder
3. Drop in an MP3 and say "make an MV for this track"
4. Claude reads the bundled CLAUDE.md, picks a template, builds the HTML, and opens it in the browser

About accuracy (chord detection)

BTC's official Weighted Chord Symbol Recall (WCSR) scores are below. Training and evaluation data is Western pop (Beatles / Queen / Robbie Williams / UsPop2002 etc.).

• Root only — 83.8%
• Maj-min — 82.7%
• Triads — 75.9%
• Sevenths — 71.8%
• Tetrads (with tensions) — 65.5%
• MIREX score — 80.8%

It's not perfect. Assume roughly 1 in 4–5 detections is a miss or ambiguous. J-Pop, anime songs, vocaloid, ambient, and electronic music are outside the training distribution, so accuracy drops further on those. In practice we recommend asking Claude to correct the generated chords.json ("the chord here is wrong, change it to E"), or using the bundled ChordCoT prompt set to refine progressions with an LLM. Details in docs/FAQ.md inside the ZIP.

Requirements

• Recommended: Anthropic Claude Pro or Max plan with Claude Code
• macOS / Linux. Windows is unverified — Windows via WSL2 is expected to work; native Windows support is not currently provided.
• Python 3.10+, FFmpeg, Node.js + Puppeteer

Reference work

Every track Emocute released in 2026 was made with this toolkit. → https://open.spotify.com/intl-en/artist/6wMTtHwTnDQlnHwHD4AizK

Cost comparison

Neural Frames

Custom MV freelance

~$30–300 / track → scales linearly

5-year cost
$1,140–5,940
Emocute Visual	$99 once — source included, modify freely, infinite renders

Price

$79 USD, one-time. Source included. Modifications allowed. No render limits.

Note on AI generation

Source code and documentation were written with Claude (LLM) assistance. The bundled sample track was generated with Suno. The tool's runtime outputs (chord detection, WebGL rendering) are deterministic — no LLM/diffusion calls during use.

Support

Read the bundled documentation first (MANUAL.html, docs/FAQ.md, memory/MEMORY.md). For unresolved issues, email support@emocutelab.com (more reliable than the itch.io comment thread).

License

• Source-code package (not a black box)
• Commercial use permitted (no credit line required)
• Redistribution prohibited (personal or intra-organization use only)
• Modifications are at the buyer's discretion

Localization

A Japanese-localized edition (v8.52_jp.zip) is sold on BOOTH: https://emocutesounds.booth.pm/items/8348451 — this itch.io listing ships the English edition only.