MEDIOCRE-MUSIC

Genre-bending AI music composition toolkit for LLM training dataset creation.

Create intricate, nuanced, and musically-sophisticated audio using our unique approach to LLM prompting for musical generation. Perfect for training LLM models on audio processing tasks like mixing, de-verbing, and effects processing.

Listen to Examples

Browse the full composition gallery →

The documentation site features playable audio, PDF scores, section navigation, and detailed analysis for every published composition.

Features

Genre Fusion - Combine classical and modern elements into unique hybrids
AI Composition - Generate ABC notation using Claude 3.7 Sonnet with streaming support
Sequential Expansion - LLM-driven multi-pass composition development with validation
Title Uniqueness - Automatic protection against duplicate composition titles
Format Conversion - ABC → MIDI → WAV → WebM pipeline with PDF scores
ABC Validation - Automatic syntax cleaning and segfault prevention
Audio Processing - Apply reverb, delay, distortion and more
Dataset Building - Create structured datasets for ML training
Interactive TUI - Browse compositions with playback and rating system
Publishing Pipeline - Automated asset generation for documentation site
Section Extraction - Automatic timeline markers from ABC section comments

Installation

Quick Start (CLI)

# Global installation
npm install mediocre-music -g

# Set your API key
export ANTHROPIC_API_KEY=your_key_here

Development Setup

# Clone and setup
git clone https://github.com/yourusername/mediocre.git
cd mediocre
npm install

# Create .env with your API key
echo "ANTHROPIC_API_KEY=your_key_here" > .env

Requirements

Node.js

Node.js 18+ (required)

API Keys

ANTHROPIC_API_KEY - Claude 3.7 Sonnet (required for generation)

Why Claude 3.7 Sonnet?

This project specifically requires Claude 3.7 Sonnet (claude-3-7-sonnet-20250219). Later Claude models (3.5, 4.0, Opus, etc.) have significantly regressed in ABC notation generation capabilities:

Claude 3.7 Sonnet excels at:

Generating syntactically correct ABC notation with proper header sequences
Understanding musical structure, voice leading, and harmonic relationships
Producing creative genre fusion that respects both classical and modern traditions
Working within abc2midi's extension syntax (%%MIDI commands, multi-voice arrangements)
Thinking compositionally rather than just textually

Later models fail at:

ABC syntax correctness (missing headers, malformed voice declarations)
Musical coherence (random notes vs. actual compositional logic)
Genre-appropriate instrumentation and rhythmic patterns
Understanding the relationship between notation and sonic output

This isn't about "newer = better." Claude 3.7 Sonnet has specific training or fine-tuning that makes it exceptional for musical notation tasks. The model is hardcoded in src/utils/claude.js - do not change it unless you want broken compositions.

External Tools

Tool	Purpose	Installation
`abcmidi`	ABC ↔ MIDI conversion	`apt install abcmidi` / `brew install abcmidi`
`abcm2ps`	ABC → PostScript	`apt install abcm2ps` / `brew install abcm2ps`
`ghostscript`	PostScript → PDF	`apt install ghostscript` / `brew install ghostscript`
`timidity`	MIDI → WAV synthesis	`apt install timidity` / `brew install timidity`
`fluidsynth`	Alternative MIDI synthesis	`apt install fluidsynth` / `brew install fluidsynth`
`sox`	Audio effects processing	`apt install sox` / `brew install sox`
`ffmpeg`	WAV → WebM encoding	`apt install ffmpeg` / `brew install ffmpeg`

NixOS

For NixOS users, you can use a shell with all dependencies:

nix-shell -p abcmidi abcm2ps ghostscript timidity fluidsynth sox ffmpeg

Quick Start

# Generate something actually interesting
mediocre generate \
  -s "Alien Transmission From a Dying Star" \
  -C "stockhausen,xenakis,ligeti,spectralism,musique concrete" \
  -M "autechre,aphex twin,boards of canada,tim hecker" \
  --producer "Brian Eno" \
  --record-label "Warp" \
  --sequential --stream-text

# Or go full chaos mode
mediocre generate \
  -C "serialism,webern,babbitt,total serialism,darmstadt" \
  -M "venetian snares,igorrr,speedcore,breakcore,merzbow" \
  -c 5 \
  --producer "Arca" \
  --record-label "PAN" \
  --solo \
  --sequential --stream-text

# Validate and convert
mediocre validate-abc -i output/*.abc
mediocre convert --to wav -d ./output

# Publish to documentation site
npm run publish:composition

Core Commands

Generate Compositions

Create new compositions from hybrid genre specifications with full creative control.

# FULL POWER: Massive genre fusion with producer/label aesthetics
mediocre generate \
  -s "Experimental Freak-Out From Another Planet" \
  -C "wagner,mahler,messiaen,babbitt,webern,serialism,xenakis,john cage,julius eastman,futurism,harry partch,musique concrete" \
  -M "eugene chadbourne,fred frith,speedcore,venetian snares,ornette coleman,autechre,lightning bolt,merzbow,math rock,zach hill,sunn o)))" \
  -c 8 \
  --producer "A.G. Cook" \
  --record-label "Hyperdub" \
  --sequential --stream-text

# Hyperpop meets 20th century classical
mediocre generate \
  -s "Deconstructed Club Symphony" \
  -C "stockhausen,ligeti,penderecki,spectralism,prepared piano" \
  -M "sophie,arca,pc music,100 gecs,bladee,drain gang" \
  --producer "Danny L Harle" \
  --record-label "PC Music" \
  --instruments "Synthesizer,Prepared Piano,808,Strings" \
  --sequential --stream-text

# Noise/drone with classical gravitas
mediocre generate \
  -s "Cathedral of Feedback" \
  -C "bruckner,messiaen,arvo part,gregorian chant,organum" \
  -M "sunn o))),boris,earth,tim hecker,fennesz,william basinski" \
  --producer "Steve Albini" \
  --record-label "Southern Lord" \
  --solo \
  --sequential --stream-text

# Math rock meets serialist complexity
mediocre generate \
  -g "Serialist_x_Math_Rock" \
  --producer "Steve Albini" \
  --instruments "Guitar,Bass,Drums,Vibraphone" \
  --solo \
  --sequential --stream-text

# Simple hybrid with creative naming
mediocre generate -g "Baroque_x_Footwork" --creative-names -c 3

# Custom prompts for full control
mediocre generate -g "Opera_x_Noise" \
  --system-prompt examples/custom-system-prompt.txt \
  --user-prompt examples/noise-opera-instructions.txt

All Generate Flags:

Flag	Description
`-g, --genre`	Hybrid genre name (e.g., "Baroque_x_Techno")
`-s, --style`	Custom style/vibe description for the composition
`-C, --classical`	Comma-separated classical influences (composers, movements, techniques)
`-M, --modern`	Comma-separated modern influences (artists, genres, scenes)
`-c, --count`	Number of compositions to generate
`--producer`	Make it sound like this producer made it
`--record-label`	Make it sound like it was released on this label
`--instruments`	Force specific instruments in the output
`--solo`	Include a virtuosic solo section
`--creative-names`	Generate wild genre names instead of "X_x_Y" format
`--system-prompt`	Custom system prompt file
`--user-prompt`	Custom user prompt file
`--sequential`	Enable LLM-driven expansion mode
`--stream-text`	Use streaming API (prevents timeouts)
`--no-midi`	Skip MIDI conversion

Modify Compositions

Transform or extend existing compositions with full creative control.

# Add a producer's signature sound
mediocre modify "output/serialist_x_breakcore-*.abc" \
  -i "Add a breakdown that sounds like it was produced by Arca - glitchy, detuned, alien" \
  --producer "Arca" \
  --sequential --stream-text

# Extend with specific instrumentation
mediocre modify "composition.abc" \
  -i "Add a 32-bar prepared piano solo in the style of John Cage, then bring back all instruments for a massive climax" \
  --instruments "Prepared Piano,Strings,Percussion" \
  --solo \
  --sequential --stream-text

# Transform the vibe entirely
mediocre modify "baroque_piece.abc" \
  -i "Make it sound like a lost recording from a 1970s Italian horror film soundtrack" \
  --producer "Goblin" \
  --record-label "Cinevox" \
  --sequential --stream-text

# Instructions from file for complex modifications
mediocre modify "piece.abc" \
  -f instructions/add-microtonal-section.txt \
  --sequential --stream-text

All Modify Flags:

Flag	Description
`-i, --instructions`	Text instructions for modification
`-f, --instructions-file`	File containing modification instructions
`--producer`	Apply producer's aesthetic to modifications
`--record-label`	Apply label's sonic identity
`--instruments`	Constrain to specific instruments
`--solo`	Add a solo section
`--sequential`	Enable validation loop
`--stream-text`	Use streaming API

Combine Compositions

Merge multiple compositions with intelligent integration.

# Combine with date and genre filtering
mediocre combine \
  -g "serialist,noise,experimental" \
  -f 2024-01-01 \
  -l 120 \
  --producer "Rashad Becker" \
  --record-label "PAN" \
  --sequential --stream-text

# Force specific instrumentation in the combined piece
mediocre combine \
  -f "drone_piece.abc" "breakcore_piece.abc" "orchestral_piece.abc" \
  --instruments "Synthesizer,Orchestra,Breakbeats" \
  --producer "Oneohtrix Point Never" \
  --sequential --stream-text

# Combine recent pieces under 60 seconds
mediocre combine \
  -l 60 \
  -f 2024-06-01 \
  --solo \
  --sequential --stream-text

All Combine Flags:

Flag	Description
`-f, --files`	Specific ABC files to combine
`-g, --genres`	Filter by genres (comma-separated)
`-l, --duration-limit`	Max duration in seconds for source pieces
`-f, --date-from`	Filter pieces created after this date
`-t, --date-to`	Filter pieces created before this date
`-d, --directory`	Directory to search for compositions
`--producer`	Apply producer aesthetic
`--record-label`	Apply label aesthetic
`--instruments`	Force specific instruments
`--solo`	Include solo section
`--sequential`	Enable validation loop
`--stream-text`	Use streaming API

Validate ABC Notation

Validate and fix ABC notation files to prevent synthesis errors.

# Validate single file
mediocre validate-abc -i "/path/to/composition.abc"

# Validate with output to new file
mediocre validate-abc -i "/path/to/composition.abc" -o "/path/to/fixed.abc"

# Validate all ABC files in output directory
mediocre validate-abc

The validator:

Checks for required headers (X:, T:, M:, L:, K:)
Ensures blank lines between header and body
Removes problematic characters that cause abc2midi segfaults
Tests compilation with abc2midi

Sanitize Drums

Remap problematic percussion notes to safe alternatives.

# Basic sanitization with standard remapping
mediocre sanitize "output/composition.abc"

# Preview what would change without modifying
mediocre sanitize "output/composition.abc" --dry-run

# LLM-powered intelligent replacement (for non-standard drum programs)
mediocre sanitize "output/composition.abc" --llm

What gets remapped:

Banned Note	Replacement	Reason
Note 71 (Whistle)	Note 45 (Low Tom)	Harsh, piercing sound
Note 62 (Mute Hi Conga)	Note 47 (Low-Mid Tom)	Poor soundfont samples
Note 63 (Open Hi Conga)	Note 48 (Hi-Mid Tom)	Inconsistent rendering
Note 73 (Short Guiro)	Note 50 (High Tom)	Glitchy artifacts
Note 74 (Long Guiro)	Note 43 (High Floor Tom)	Rendering issues
Note 78 (Mute Cuica)	Note 41 (Low Floor Tom)	Unpleasant timbre
Note 79 (Open Cuica)	Note 45 (Low Tom)	Unreliable synthesis

Flags:

Flag	Description
`--dry-run`	Preview changes without modifying files
`--llm`	Use Claude for intelligent context-aware replacement

More Like This

Generate compositions similar to an existing piece.

# Create variations on an existing composition
mediocre more-like-this "output/serialist_x_noise-masterpiece.abc" -c 3

# With producer/label aesthetic applied to variations
mediocre more-like-this "output/baroque_x_footwork.abc" \
  -c 5 \
  --producer "Flying Lotus" \
  --record-label "Brainfeeder" \
  --solo

# Apply a different style to variations
mediocre more-like-this "output/impressionist_x_ambient.abc" \
  -s "Darker, more industrial" \
  --creative-names

Mix and Match

Combine elements from multiple compositions into something new.

# Basic mix and match
mediocre mix-and-match \
  -f "output/piece1.abc" "output/piece2.abc" "output/piece3.abc"

# With specific instrumentation and production
mediocre mix-and-match \
  -f "baroque.abc" "noise.abc" "ambient.abc" \
  --instruments "Harpsichord,Synthesizer,Feedback,Strings" \
  --producer "Ben Frost" \
  --record-label "Mute" \
  --solo

Add Lyrics

Add vocal parts to existing compositions.

# Basic lyrics addition
mediocre lyrics \
  -m "output/composition.mid" \
  -a "output/composition.abc" \
  -p "A song about algorithmic entropy and digital decay"

# With full production control
mediocre lyrics \
  -m "output/opera_x_hyperpop.mid" \
  -a "output/opera_x_hyperpop.abc" \
  -p "An aria about parasocial relationships and internet fame" \
  --producer "SOPHIE" \
  --record-label "Transgressive" \
  --instruments "Voice,Synthesizer,808,Strings"

Advanced Features

Sequential Expansion Mode

The --sequential flag enables LLM-driven multi-pass composition development. Instead of generating a complete piece in one shot, it:

Generates an initial composition framework
Validates with abc2midi to catch syntax errors
Evaluates completeness using genre-aware criteria
Expands iteratively until the piece is musically complete

# Enable sequential expansion
mediocre generate -g "Minimalist_x_Drum_and_Bass" --sequential

How it works:

Initial Generation
       ↓
abc2midi Validation ←──── Fix Errors (if needed)
       ↓
Completeness Evaluation
       ↓
   Complete? ─── Yes ──→ Done
       ↓
      No
       ↓
   Expansion Pass ───→ abc2midi Validation ───→ Loop

The system uses Claude to evaluate whether a composition needs expansion based on:

Genre-specific completeness criteria
Musical structure (intro, development, conclusion)
Voice/instrument utilization
Rhythmic and harmonic development

Safety limits: Maximum 10 passes to prevent infinite loops.

Streaming Mode

The --stream-text flag uses the Vercel AI SDK's streamText for generation, which:

Prevents timeouts on large compositions
Shows progress during generation
Handles long pieces that would otherwise fail

# Always use with sequential for best results
mediocre generate -g "Opera_x_Vaporwave" --sequential --stream-text

When to use streaming:

Compositions with many voices/instruments
Sequential expansion mode (multiple LLM calls)
Long-form pieces (symphonies, operas)
Any generation that times out without it

Title Uniqueness Protection

Every generated composition automatically receives a unique title. The system:

Loads existing titles from docs/data/compositions.json
Checks new titles against the database (case-insensitive)
Regenerates duplicate titles using Claude Haiku (fast, cheap)
Falls back to timestamp-based uniqueness if needed

This prevents duplicate titles like multiple "SERIALIST CHAOS" or "PREPARED NOISE" compositions.

No configuration needed - this runs automatically on every generation.

ABC Validation Pipeline

The validation pipeline prevents common synthesis failures:

// Automatic cleaning applied to all generated ABC
cleanAbcNotation(abc); // Removes segfault-causing characters
validateAbcNotation(abc); // Checks structure and headers
validateWithAbc2Midi(file); // Tests actual compilation

What gets cleaned:

Unicode characters that crash abc2midi
Missing blank lines between header and body
Invalid header sequences
Problematic accidentals and ornaments

Publishing Pipeline

The publishing system converts compositions to web-ready assets for the documentation site.

# Interactive selection of unpublished compositions
npm run publish:composition

# List all compositions (published and unpublished)
npm run publish:composition -- --list

# Unpublish a composition
npm run publish:composition -- --unpublish <composition-id>

# Dry run (show what would happen)
npm run publish:composition -- --dry-run

Pipeline Steps

Select unpublished composition from ABC files
Validate ABC notation with abc2midi
Generate MIDI from ABC notation
Synthesize WAV using TiMidity with custom soundfonts
Encode WebM with ffmpeg (Opus audio, web-optimized)
Generate PDF score using abcm2ps + ghostscript
Extract Sections from ABC comments for timeline markers
Update compositions.json with metadata

Asset Outputs

Asset	Location	Purpose
WebM audio	`docs/media/<id>.webm`	Web playback
PDF score	`docs/media/<id>-score.pdf`	Sheet music
Metadata	`docs/data/compositions.json`	Gallery data

Section Extraction

The pipeline extracts section markers from ABC comments:

% Section: Introduction
V:1
...music...

% Section: Development
V:1
...music...

Or numbered format:

% 1. Exposition
...music...

% 2. Development
...music...

Sections include calculated start times based on tempo, meter, and measure counts.

Documentation Site

The docs/ directory contains an interactive composition gallery.

Structure

docs/
├── index.html          # Main page
├── data/
│   └── compositions.json  # Composition database
├── media/              # Audio and PDF assets
├── js/
│   ├── app.js          # Main application
│   └── components/
│       ├── music-player.js    # Audio player with sections
│       ├── music-card.js      # Composition cards
│       └── music-gallery.js   # Gallery grid
└── css/
    └── styles.css      # Terminal-style theme

Features

Genre Filtering - Filter by classical or modern genre tags
Sorting - Sort by date, title, or genre
Modal View - Detailed composition info with analysis
Section Links - Click "Section I" in analysis to seek audio
PDF Viewer - View sheet music in modal
Responsive - Works on mobile and desktop

Running Locally

# Simple HTTP server
cd docs
python -m http.server 8000
# or
npx serve .

TiMidity Configuration

The timidity-sanitized.cfg provides high-quality MIDI synthesis with 30+ layered soundfonts.

Soundfont Strategy

Layer 1: Base GM/GS/XG banks (GeneralUser GS, FluidR3)
Layer 2: Specialized instruments (orchestral, synths)
Layer 3: Premium overrides (Orpheus, Crisis, Musica)
Layer 4: Drum kits (jazz, rock, electronic)

Key Features

Instrument Overrides:

# Better trumpet sound
bank 0
56 %font "Airfont_340.sf2" 0 56  # Trumpet override

Drum Note Remapping: Problematic percussion sounds are remapped to safe alternatives:

Original	Remapped To	Reason
Note 71 (Whistle)	Note 45 (Low Tom)	Harsh sound
Note 62 (Mute Hi Conga)	Note 47 (Low-Mid Tom)	Poor sample
Note 73 (Guiro)	Note 50 (High Tom)	Rendering issues

Drumset Coverage:

Standard Kit (0)
Room Kit (8)
Power Kit (16)
TR-808 Kit (25)
TR-606 Kit (28)
Jazz Kit (32)
Brush Kit (40)
Orchestra Kit (48)
SFX Kit (56)

Using Custom Config

# With timidity directly
timidity -c timidity-sanitized.cfg input.mid -Ow -o output.wav

# With fluidsynth (alternative)
fluidsynth -ni /path/to/soundfont.sf2 input.mid -F output.wav

Hybrid Genre System

The -C (classical) and -M (modern) flags accept anything - genres, composers, artists, movements, techniques, aesthetics. Go wild.

Suggested Classical Influences (`-C`)

Composers:

Bach, Mozart, Beethoven, Brahms, Wagner, Mahler, Bruckner
Debussy, Ravel, Satie, Fauré (Impressionist)
Schoenberg, Webern, Berg, Babbitt (Serialist/12-tone)
Stockhausen, Xenakis, Ligeti, Penderecki (Post-war avant-garde)
Messiaen, Boulez, Berio, Nono (Spectralist/Modernist)
Reich, Glass, Riley, Adams (Minimalist)
Cage, Feldman, Tudor, Wolff (Experimental/Chance)
Partch, Johnston, Tenney (Microtonal)
Julius Eastman, Pauline Oliveros, Alvin Lucier
Arvo Pärt, Górecki, Tavener (Holy Minimalism)

Movements/Techniques:

Serialism, Spectralism, Minimalism, Maximalism
Musique Concrète, Aleatory, Prepared Piano
Futurism, Dadaism, Fluxus
Total Serialism, Stochastic Music
Extended Techniques, Graphic Scores

Forms:

Opera, Cantata, Oratorio, Mass
Symphony, Concerto, Sonata, Fugue
Chamber Music, String Quartet
Choral, A Cappella, Gregorian Chant

Suggested Modern Influences (`-M`)

Artists/Bands:

Autechre, Aphex Twin, Boards of Canada, Squarepusher (IDM)
Venetian Snares, Igorrr, Machine Girl (Breakcore)
Merzbow, Masami Akita, Prurient, Pharmakon (Noise)
Sunn O))), Earth, Boris, Khanate (Drone/Doom)
Lightning Bolt, Hella, Don Caballero, Battles (Math Rock)
SOPHIE, Arca, A.G. Cook, Danny L Harle (Hyperpop/PC Music)
100 Gecs, Bladee, Drain Gang, Yung Lean
Flying Lotus, Thundercat, Kamasi Washington (Brainfeeder)
Tim Hecker, Fennesz, William Basinski (Ambient/Drone)
Oneohtrix Point Never, James Ferraro (Vaporwave/Hypnagogic)
Fred Frith, Eugene Chadbourne, John Zorn (Improv/Avant-Rock)
Zach Hill, Death Grips, clipping. (Experimental Hip-Hop)
Ornette Coleman, Sun Ra, Albert Ayler (Free Jazz)

Genres:

Speedcore, Gabber, Hardcore, Terrorcore
Footwork, Juke, Ghettotech, Ballroom
Grime, Drill, UK Drill, Trap
Vaporwave, Mallsoft, Future Funk
Noise, Power Electronics, Harsh Noise Wall
Black Metal, Death Metal, Grindcore
Post-Punk, No Wave, Industrial

Suggested Record Labels (`--record-label`)

The label aesthetic shapes the entire vibe:

Label	Aesthetic
Hyperdub	Bass-heavy, UK garage roots, experimental club
Warp	IDM, abstract electronics, boundary-pushing
PC Music	Hypergloss, maximalist pop, uncanny valley
PAN	Experimental, deconstructed club, avant-garde
Brainfeeder	Jazz fusion, cosmic funk, beat science
Southern Lord	Drone, doom, crushing heaviness
Mute	Industrial, electronic body music, dark synth
4AD	Ethereal, dreamy, shoegaze aesthetics
Kranky	Drone, ambient, patient soundscapes
Raster-Noton	Glitch, microsound, precise digital
Editions Mego	Noise, computer music, experimental
Thrill Jockey	Post-rock, experimental, Chicago sound
Sub Pop	Grunge, indie, Pacific Northwest
Ninja Tune	Trip-hop, breakbeat, DJ culture
Planet Mu	Footwork, IDM, leftfield electronics

Suggested Producers (`--producer`)

The producer shapes the sonic signature:

Producer	Signature
A.G. Cook	Hypercompressed, pitch-shifted, saccharine
Arca	Glitchy, detuned, alien, fluid
SOPHIE	Latex textures, hard surfaces, synthetic
Flying Lotus	Cosmic, jazz-inflected, beat collage
Oneohtrix Point Never	Nostalgic, processed, uncanny
Steve Albini	Raw, dynamic, room sound
Brian Eno	Ambient, generative, textural
Burial	Crackle, distance, 2-step rhythms
Rashad Becker	Sculptural, detailed, microsound
Ben Frost	Harsh, cinematic, overwhelming
Goblin	Horror, Italian prog, tension
Vangelis	Synth pads, sci-fi, emotional
Timbaland	Syncopated, unexpected, pop-forward
Pharrell	Minimal, bouncy, clean
Metro Boomin	Dark, atmospheric, 808-driven

Example Commands

# John Zorn meets PC Music
mediocre generate \
  -s "Avant-Pop Freakout" \
  -C "john zorn,naked city,tzadik,game piece,file card composition" \
  -M "100 gecs,sophie,a.g. cook,hyperpop,nightcore" \
  --producer "A.G. Cook" \
  --record-label "PC Music" \
  --sequential --stream-text

# Black metal meets spectralism
mediocre generate \
  -s "Ritual of Frequency" \
  -C "grisey,murail,spectralism,ligeti,scelsi,xenakis" \
  -M "deathspell omega,blut aus nord,ulver,sunn o))),liturgy" \
  --producer "Colin Marston" \
  --record-label "Profound Lore" \
  --sequential --stream-text

# Footwork meets minimalism
mediocre generate \
  -s "160 BPM Gradual Process" \
  -C "steve reich,terry riley,la monte young,philip glass,in c" \
  -M "rp boo,dj rashad,dj spinn,jlin,footwork,teklife" \
  --producer "Flying Lotus" \
  --record-label "Planet Mu" \
  --instruments "808,Marimba,Synthesizer,Sampler" \
  --sequential --stream-text

# Free jazz meets harsh noise
mediocre generate \
  -s "Total Freedom" \
  -C "ornette coleman,free jazz,sun ra,albert ayler,cecil taylor,aacm" \
  -M "merzbow,prurient,pharmakon,kevin drumm,harsh noise wall" \
  --producer "John Zorn" \
  --record-label "Tzadik" \
  --solo \
  --sequential --stream-text

CLI Reference

All Commands

Command	Description
`generate`	Create new compositions
`modify`	Transform existing compositions
`combine`	Merge multiple compositions
`genres`	Generate hybrid genre names
`convert`	Convert ABC to MIDI/WAV/PDF
`process`	Apply audio effects
`validate-abc`	Validate and fix ABC notation
`sanitize`	Remap problematic drum notes
`info`	Display composition information
`more-like-this`	Generate similar compositions
`mix-and-match`	Combine elements from multiple files
`lyrics`	Add lyrics to compositions
`dataset`	Build ML training datasets
`tui`	Interactive terminal browser

Global Flags

Flag	Description
`--sequential`	Enable LLM-driven expansion mode
`--stream-text`	Use streaming API
`--midi` / `--no-midi`	Control MIDI generation
`-d, --output-dir`	Output directory (default: ./output)
`--verbose`	Enable verbose logging

Tips & Best Practices

For Best Results

Always use --sequential --stream-text for complex genres

mediocre generate -g "Spectralist_x_IDM" --sequential --stream-text

Validate before converting to catch issues early
```
mediocre validate-abc -i output/*.abc
```
Use the publish pipeline for production-ready assets
```
npm run publish:composition
```
Layer soundfonts in TiMidity for richer sound
- Base: GeneralUser GS (comprehensive)
- Orchestral: Sonatina Symphony Orchestra
- Synths: Premium soundfonts for electronic genres

Genre Selection Tips

Serialist + Electronic genres work well (structured chaos)
Impressionist + Ambient creates beautiful soundscapes
Baroque + Modern provides strong rhythmic contrast
Prepared Piano + Noise for experimental textures

Performance Tips

Streaming mode prevents API timeouts on long pieces
Sequential mode produces more complete compositions
Batch validation catches issues before synthesis
Custom soundfonts dramatically improve audio quality

Common Workflows

New Composition:

mediocre generate -g "Genre_x_Genre" --sequential --stream-text
mediocre validate-abc -i output/genre_x_genre-*.abc
npm run publish:composition

Extend Existing:

mediocre modify "composition.abc" -i "Add virtuosic coda" --sequential --stream-text

Combine Pieces:

mediocre combine -f "piece1.abc" "piece2.abc" --sequential --stream-text

Troubleshooting

Debug Mode

# Verbose output for troubleshooting
mediocre generate -g "Test_x_Test" --verbose

# Check abc2midi directly
abc2midi input.abc -o output.mid 2>&1 | head -50

Getting Help

# Command help
mediocre --help
mediocre generate --help

# Check version
mediocre --version

License

MIT

Made with mediocrity

Name		Name	Last commit message	Last commit date
Latest commit History 41 Commits
dist		dist
docs		docs
examples		examples
scripts		scripts
src		src
tests		tests
.gitignore		.gitignore
README.md		README.md
fluidsynth-play.sh		fluidsynth-play.sh
fluidsynth-render.sh		fluidsynth-render.sh
jest.config.js		jest.config.js
jsconfig.json		jsconfig.json
package-lock.json		package-lock.json
package.json		package.json
screenshot.png		screenshot.png
timidity-sanitized.cfg		timidity-sanitized.cfg
timidity-test-config.cfg		timidity-test-config.cfg
timidity.cfg		timidity.cfg

agrathwohl/mediocre

Folders and files

Latest commit

History

Repository files navigation

MEDIOCRE-MUSIC

Listen to Examples

Table of Contents

Features

Installation

Quick Start (CLI)

Development Setup

Requirements

Node.js

API Keys

Why Claude 3.7 Sonnet?

External Tools

NixOS

Quick Start

Core Commands

Generate Compositions

Modify Compositions

Combine Compositions

Validate ABC Notation

Sanitize Drums

More Like This

Mix and Match

Add Lyrics

Advanced Features

Sequential Expansion Mode

Streaming Mode

Title Uniqueness Protection

ABC Validation Pipeline

Publishing Pipeline

Pipeline Steps

Asset Outputs

Section Extraction

Documentation Site

Structure

Features

Running Locally

TiMidity Configuration

Soundfont Strategy

Key Features

Using Custom Config

Hybrid Genre System

Suggested Classical Influences (-C)

Suggested Modern Influences (-M)

Suggested Record Labels (--record-label)

Suggested Producers (--producer)

Example Commands

CLI Reference

All Commands

Global Flags

Tips & Best Practices

For Best Results

Genre Selection Tips

Performance Tips

Common Workflows

Troubleshooting

Debug Mode

Getting Help

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Languages

Suggested Classical Influences (`-C`)

Suggested Modern Influences (`-M`)

Suggested Record Labels (`--record-label`)

Suggested Producers (`--producer`)

Packages