MIDI Sketch is a music theory-based MIDI generator that creates complete pop music arrangements. Unlike AI audio generators like Suno, it outputs editable MIDI data - drums, bass, chords, arpeggios, and melodies - that you can import into your DAW and customize with your own sounds and mixing.

How is MIDI Sketch different from Suno or other AI music generators?

AI audio generators like Suno produce finished audio that you cannot edit. MIDI Sketch generates MIDI data based on music theory, giving you full control. You can change instruments, edit notes, adjust timing, and mix however you want in your DAW. Same seed always produces the same output - fully reproducible.

How do I use MIDI Sketch with my DAW?

Generate your song in the browser, then download the MIDI file. Import it directly into any DAW like Ableton Live, FL Studio, Logic Pro, or Cubase. All tracks are separated and ready for your own sounds and mixing.

What music styles does MIDI Sketch support?

MIDI Sketch includes presets for J-Pop, K-Pop, City Pop, EDM, Ballad, Rock, R&B, and more. Each preset configures appropriate rhythms, chord voicings, and arrangement patterns based on music theory for that style.

How are chord progressions generated?

MIDI Sketch auto-generates chord progressions based on popular patterns in each genre. Select a style preset and the system creates appropriate progressions with jazz voicings, tensions, and extensions - all grounded in music theory.

Is MIDI Sketch output reproducible?

Yes, MIDI Sketch is fully deterministic. The same seed and settings always produce exactly the same MIDI output. This makes it perfect for iterative production workflows where you need consistent, predictable results.

How is the melody generated?

MIDI Sketch generates melodies based on music theory - following chord tones, using appropriate scales, and creating natural phrase structures. You can configure vocal ranges and regenerate with different seeds until you find a melody you like.

What tracks are included in the MIDI output?

A complete MIDI Sketch export includes: drums (kick, snare, hi-hat, fills), bass line, chord pads, optional arpeggio, and vocal melody. Each track is on a separate MIDI channel for easy DAW editing.

Can I use MIDI Sketch for commercial music production?

Yes, MIDI Sketch is MIT licensed. All generated MIDI files are yours to use commercially. Use them as starting points for your productions, demos, or final releases.

Why use MIDI instead of AI-generated audio?

MIDI gives you complete creative control. You can change every note, swap instruments, adjust velocities, quantize timing, and mix with your own effects. AI audio is a black box - MIDI is transparent and editable. For serious music production, MIDI is the professional choice.

Features

MIDI Sketch is a music theory-based MIDI generator that creates complete pop music arrangements.

MIDI Output, Not Audio

Unlike AI audio generators (Suno, Udio, etc.), MIDI Sketch outputs editable MIDI data.

	AI Audio Generators	MIDI Sketch
Output	Finished audio (MP3/WAV)	MIDI files
Editing	Limited or none	Full control in DAW
Sounds	Fixed	Your choice
Mixing	Baked in	You decide
Reproducibility	Often inconsistent	Deterministic (seed-based)

What You Get

9 separate tracks (vocal, aux, chord, bass, motif, guitar, arpeggio, drums, SE)
Each track on its own MIDI channel
Import directly into any DAW
Use your own instruments and effects

Music Theory Foundation

MIDI Sketch doesn't use machine learning or neural networks. It implements classical harmony principles combined with modern pop music analysis.

Melody Generation

Template-Driven Architecture

7 melody templates model specific vocal styles:

PlateauTalk: NewJeans/Billie Eilish style - high plateau with talk-sing
RunUpTarget: YOASOBI/Ado style - ascending runs to target notes
HookRepeat: TikTok/K-POP style - short repeating hooks
SparseAnchor: Official髭男dism style - sparse anchor notes
And more (DownResolve, CallResponse, JumpAccent)

Singability Constraints

Direction inertia: Accumulated momentum tracking prevents erratic direction changes
Tessitura enforcement: Real-time pitch adjustments for comfortable singing range
Leap compensation: Automatic stabilization steps after large intervals
Vowel constraints: Pitch movement limited within vowel sections for natural phrasing

Voice Leading & Chord Voicing

Three Voicing Types

Close voicing: Notes within one octave (warm, suitable for verses)
Open voicing: Drop2, Drop3, Spread variations (powerful, for choruses)
Rootless voicing: Root omitted when bass provides it (jazz-influenced)

Voice Leading Optimization

Weighted distance calculation (bass and soprano get 2x priority)
Common tone maximization between successive chords
Parallel 5ths/octaves detection with context-aware enforcement
Avoid note detection (minor 2nd with chord tones, tritone with root)

Non-Chord Tone (NCT) Decoration

Based on Kostka & Payne's Tonal Harmony framework:

Strong Beats and Weak Beats

In 4/4 time, strong beats (1 and 3) feel accented and stable, while weak beats (2 and 4) feel lighter. Melodies typically place chord tones on strong beats for harmonic clarity.

NCT Types

Type	Placement	Description
Passing Tone	Weak beat	Stepwise connection between chord tones
Neighbor Tone	Weak beat	Step away from chord tone and return
Appoggiatura	Strong beat	Accented dissonance resolving by step
Anticipation	Before beat	Early arrival of next chord tone
Tension	Context-dependent	9th, 11th, 13th extensions

Mood-Dependent Configuration

Bright/Upbeat: 75% chord tones, pentatonic focus
CityPop: 50% chord tones, jazz tensions enabled
Ballad: 65% chord tones, expressive appoggiaturas
Dark/Dramatic: Chromatic approach notes enabled

Harmony Context & Collision Avoidance

Multi-Track Coordination

Track collision detection: Registers all notes from vocal, bass, chord, aux tracks
Low register strictness: 3-semitone threshold below C4 to prevent muddiness
Safe pitch resolution: Multi-strategy fallback (chord tones → consonant intervals → range search)

Emotion Curve System

Song Emotional Arc

The Emotion Curve system plans the emotional journey of a song, assigning specific characteristics to each section:

Intro: Anticipation (low tension, building energy)
Verse (A): Expectation (moderate tension)
Pre-chorus (B): Tension build (high tension, upward pitch tendency)
Chorus: Release/resolution (peak energy, maximum density)
Bridge: Reflection (lower energy, contrast)
Outro: Closure (decreasing tension)

Each section receives emotion parameters (tension, energy, resolution need, pitch tendency, density) that guide generation across all tracks.

Euclidean Rhythms

Mathematical Rhythm Patterns

Drum patterns use Bjorklund's algorithm to distribute hits evenly across steps, creating natural-sounding rhythms found in many musical traditions:

Pattern	Hits/Steps	Traditional Name
E(3,8)	[x..x..x.]	Cuban tresillo
E(5,8)	[x.xx.xx.]	Cuban cinquillo
E(5,16)	Bossa nova feel	-
E(4,16)	Four-on-the-floor	-

These mathematically-spaced patterns feel more natural than probability-based random placement.

Secondary Dominants

Harmonic Enrichment

Secondary dominants (V/V, V/vi, etc.) are automatically inserted to create stronger harmonic pull toward target chords. This enriches chord progressions without requiring manual configuration.

Guitar Track

Accompaniment Guitar Generation

A dedicated guitar track generates accompaniment patterns influenced by Blueprint constraints such as guitar skill level and guitar-below-vocal positioning. Guitar appears on its own MIDI channel and can be enabled or disabled independently.

Energy Curve

Song Energy Progression

The Energy Curve system controls how energy progresses through the song, providing high-level control over dynamics beyond per-section settings:

GradualBuild: Energy increases steadily from start to finish
FrontLoaded: High energy at the start, tapering toward the end
WavePattern: Alternating high and low energy across sections
SteadyState: Consistent energy level throughout

Melody & Motif Overrides

Fine-Grained Parameter Control

Melody Override allows fine-grained control over melody generation parameters:

Max leap, syncopation probability, phrase length
Long note ratio, chorus register shift
Hook repetition, leading tone behavior

Motif Override allows fine-grained control over motif generation parameters:

Motif length, note count, motion (0-4)
Register (high/mid), rhythm density

Expanded Arpeggio Patterns

8 Arpeggio Patterns

Beyond the basic Up, Down, UpDown, and Random patterns, MIDI Sketch now includes:

Pinwheel: Alternating direction pattern
PedalRoot: Returns to root between each note
Alberti: Classical broken chord pattern (low-high-mid-high)
BrokenChord: Irregular chord tone ordering

Performance Controls

DriveFeel, Syncopation & Mora Rhythm

DriveFeel: Controls performance intensity from laid-back (0) to aggressive (100), affecting timing tightness and velocity emphasis
Syncopation: enableSyncopation toggle adds groove effects by shifting notes off the grid
MoraRhythmMode: Support for Japanese mora-timed rhythm, aligning note durations to syllable timing patterns

Piano Roll Safety API

Note Safety Analysis

The Piano Roll Safety API analyzes pitch safety at any point in the generated song. For each MIDI pitch (0-127), it reports:

Safety level: Safe (chord tone), Warning (tension/low register/passing tone), or Dissonant (non-scale/collision)
Reason flags: Detailed bit flags indicating why a pitch is rated at its level (e.g., ChordTone, Tension, LargeLeap, Minor2nd collision)
Collision detection: Identifies which tracks would collide with a given pitch
Recommended pitches: Up to 8 suggested pitches for the current harmonic context

Use after any generation call (generateVocal, generateFromConfig, etc.) for real-time pitch guidance in piano roll editors.

Custom Vocal API

User-Defined Melody Input

The setVocalNotes API allows injecting a custom melody (as an array of note events) instead of using the built-in melody generator. The accompaniment is then generated around the user-provided vocal, with full harmony context coordination including chord recognition, collision avoidance, and Aux track generation.

Chord Timeline API

Harmonic Context Retrieval

The getChordTimeline API returns the chord progression timeline for the generated song, including tick positions, chord degrees, and secondary dominant information. This is used for playback synchronization and harmonic analysis.

SongConfigBuilder

Fluent Configuration API

The SongConfigBuilder provides a fluent API for constructing song configurations with cascade change detection. When a parameter changes, dependent parameters are automatically recalculated, ensuring consistent configurations without manual coordination.

Academic Foundation

The implementation references:

Kostka & Payne: Tonal Harmony - NCT classification and voice leading
Huron: Sweet Anticipation - Psychology of musical expectation
de Clercq & Temperley: A Corpus Analysis of Rock Harmony - Pop/rock chord progression patterns
J-POP pentatonic "yonanuki" analysis

Deterministic Generation

Same seed + same parameters = same output. Every time.

bash

# These will always produce identical MIDI files
./midisketch_cli --seed 12345 --style jpop
./midisketch_cli --seed 12345 --style jpop

Reproducibility Benefits

Reproducible results for iterative workflows
Share seeds with collaborators
Metadata embedded in MIDI files enables regeneration

Candidate Selection System

For melody generation, MIDI Sketch doesn't just output the first result. It generates 20-100 candidates per section and selects the best one through evaluation:

Culling: Filter out melodies with issues (high register strain, monotony, scattered notes)
Scoring: Rank survivors on singability, chord tone alignment, contour shape
Selection: Choose the highest-scoring candidate

Candidate Counts by Section

Section	Candidates
Chorus	100
Pre-chorus (B)	50
Bridge / Chant	30
Verse / Intro / Outro	20

More candidates for important sections where melody quality matters most.

Style Presets

17 style presets (stylePresetId 0-16) determine the overall musical character, each mapped to one of 24 internal moods. Moods (0-23) can also be set explicitly via moodExplicit, covering:

J-Pop / K-Pop / City Pop
EDM / Electro Pop / Synthwave / Future Bass
Ballad / R&B / R&B Neo Soul / Chill / Lofi
Rock / Light Rock
Anime / Vocaloid
Latin Pop / Trap
And more

What Each Preset Configures

BPM range
Drum patterns
Chord voicing style
Melody template preferences
Evaluation weights
Mood-dependent chord extension probabilities

14 vocal style presets are available (Auto, Standard, Vocaloid, UltraVocaloid, Idol, Ballad, Rock, CityPop, Anime, BrightKira, CoolSynth, CuteAffected, PowerfulShout, KPop) to fine-tune melody generation characteristics independently from mood.

Multiple Composition Styles

Three composition paradigms:

Style	Vocal	Aux	Motif	Arpeggio	Use Case
MelodyLead (0)	Yes	Yes	Blueprint-dependent	Optional	Songs with vocals
BackgroundMotif (1)	No	Yes	Yes	Optional	BGM, lo-fi
SynthDriven (2)	No	No	Blueprint-dependent	Optional (manual enable)	Electronic, EDM

BGM-Only Modes

BackgroundMotif disables Vocal but keeps Aux enabled and forces Motif generation. SynthDriven disables both Vocal and Aux; Arpeggio must be manually enabled with arpeggioEnabled=true.

Vocal-First Workflow

For MelodyLead style, iterate on the melody before generating accompaniment:

Iterate Until Satisfied

Use generateVocal() to create the initial melody, then call regenerateVocal() with a new seed or VocalConfig to try variations. Once satisfied, call generateAccompaniment() to add the backing tracks. Alternatively, use generateWithVocal() for vocal-priority one-shot generation.

Lightweight & Portable

~555KB WASM (gzip: ~225KB) + ~80KB JS
No external dependencies (pure C++17)
Runs in browser, Node.js, or native CLI
No API calls, no internet required

Open Source

License

Apache 2.0 licensed - use generated MIDI commercially, modify and redistribute freely.

Use Cases

Demo Production

Generate quick song sketches to test ideas before investing time in full production.

Learning Tool

Study how chord progressions, voice leading, and arrangement work by examining the output.

DAW Templates

Generate starting points for tracks, then customize with your own sounds and mixing.

Game/Video BGM

Create reproducible background music with deterministic seeds.

Songwriting Aid

Get melody ideas and chord progressions to build upon.

What MIDI Sketch Is Not

Important Distinctions

Not an AI audio generator - It outputs MIDI, not audio
Not a replacement for composers - It's a tool to generate starting points
Not machine learning - It uses explicit music theory rules
Not cloud-based - Everything runs locally

Features ​

MIDI Output, Not Audio ​

Music Theory Foundation ​

Melody Generation ​

Voice Leading & Chord Voicing ​

Non-Chord Tone (NCT) Decoration ​

Harmony Context & Collision Avoidance ​

Emotion Curve System ​

Euclidean Rhythms ​

Secondary Dominants ​

Guitar Track ​

Energy Curve ​

Melody & Motif Overrides ​

Expanded Arpeggio Patterns ​

Performance Controls ​

Piano Roll Safety API ​

Custom Vocal API ​

Chord Timeline API ​

SongConfigBuilder ​

Deterministic Generation ​

Candidate Selection System ​

Style Presets ​

Multiple Composition Styles ​

Vocal-First Workflow ​

Lightweight & Portable ​

Open Source ​

Use Cases ​

What MIDI Sketch Is Not ​