Generate a complete, multi-track MIDI song in your browser. No account, no server, no upload.
Pick a style, a key and a tempo: suMidi writes drums, bass, guitar, piano and ensemble parts — and
a sung melody on top — plays them back in the page, and exports a multi-track .mid file ready for
your DAW. Everything runs client-side — the music never leaves your machine.
(30-second demo coming here — meanwhile, the link above runs the real thing.)
🇮🇹 Leggi questa pagina in italiano
- A sung melody over the band, new in 2.0 — see below.
- 13 styles: MTV Unplugged, Folk Acoustic, Jazz Ballad, Neo Soul, Classical Chamber, Pop Rock, Blues Rock, Singer/Songwriter, Cinematic/Orchestral, Lo-Fi, Punk, Garage Rock, 8-bit/Chiptune.
- Session Mode: a DAW-inspired arrangement view — one lane per section, a flyout per instrument, and an interactive chord track you can edit by hand (slash chords included) with automatic voice leading.
- Solo Mode: hear a single instrument (piano, guitar, bass or ensemble) play the whole progression on its own, either in a fixed style or adapting to each section type. One click, independent from the full arrangement.
- 31 characters: every instrument — drums, drum machine, world percussion, bass, guitar, piano, ensemble — can be handed to a player with their own style and feel.
- Drum machine: a 16-step sequencer with Trap, Lo-Fi and Electro presets, editable step by step.
- Deterministic seed: same seed, style, key and tempo always produce the exact same song.
The seed is visible and pasteable in the composer bar, and the page URL carries the whole
state (
?style=&key=&bpm=&seed=) — so a song can be found again days later, or shared as a link, without anything being stored on a server. - Your work comes back: the arrangement restores itself when you reopen the page. Saving to
and reopening a
.sumidi.jsonproject file is built in, but its buttons are hidden while they are re-verified. - In-app playback through WebAudioFont — listen to one section or the whole song without exporting first.
- Export: multi-track MIDI, one instrument per channel.
- Undo up to 10 steps, with adjustable humanisation of timing and velocity.
On top of the band, suMidi writes a sung lead line — on by default, with its own lane, its own panel in each section and its own track in the MIDI file.
- Four voices — soprano, alto, tenor, baritone — chosen for the whole song in the composer bar (Voce) or section by section in the melody panel. Each sings in its own comfortable range.
- It sings like a voice: mostly stepwise, legato within a phrase, with a breath at the end of each one; choruses louder than verses; a motif that comes back, so you recognise the chorus.
- It stays in tune with the band: in the key of the song, away from the chord tones it would clash with. And the band makes room for it: piano and strings move below a soprano or an alto, and an accompanying note that would grind against a held melody note moves to the nearest chord tone.
- A timbre for each style: sax in jazz and blues, overdriven guitar in rock, oboe in classical, violin in folk and cinematic, synth lead in chiptune… or one of eight, picked by hand in the Mixer.
- Readable: the Tab & Accordi window has a Melodia tab with the melody as guitar tablature, and the chord above each bar.
There is no build step, but you do need a static server: the app uses ES modules, which
browsers refuse to load over file://.
git clone https://github.com/johnwhaledev/sumidi.git
cd sumidi
npx serve .
# then open the URL it prints (e.g. http://localhost:3000)For development (tests, linting):
npm install
npm test # regression suite (vitest)
npm run lint # ESLintindex.html the app
manual.html user manual
styles.css interface styles
fonts/ self-hosted fonts (Ubuntu, woff2)
soundfonts/ local audio presets for playback (in the repo, 18 MB)
vendor/ the WebAudioFont player library
scripts/ support scripts (e.g. download-soundfonts.mjs, impronta-note.mjs)
src/ JS modules (generation engine + UI)
SongArchitect.js builds the song structure (sections, harmony)
ChordTheory.js, ProgressioniGradi.js, SongProgressions.js, SongForms.js,
SectionPresets.js, Styles.js musical data (chords, pools, forms, styles)
*Generator.js one generator per instrument (bass, guitar, piano, drums, ensemble, pad,
melody)
SpazioVoce.js, NonPestareLaVoce.js the band makes room for the melody
Ornaments.js shared engine for glides and portamento
FlowCore.js shared utilities (RNG, dynamics, phrase memory)
Playback.js in-app playback engine (WebAudioFont)
CharacterRoster.js, GrooveLock.js, Humanizer.js characters, groove, humanisation
MidiWriter.js, TabRenderer.js, MarkdownExporter.js MIDI / tab / Markdown export
SessionManager.js, SessionStore.js, SessionExport.js, AppState.js state, saving, export
SongEngine.js whole-blueprint gen() and the composer bar
SongEngineLab.js the Classic panel, loaded only by lab.html
Session.js Session Mode: panels, chord track, playback, export, Solo Mode
main.js bootstrap
design/ UI components (DesignSystem.js) and design material
tests/ regression suite (vitest)
img/ instrument and character icons, README screenshots
Built for up-to-date desktop browsers (Chrome, Edge, Firefox).
Nothing is fetched from the network. The WebAudioFont
library and the 70 playback presets live in the repository (vendor/ and soundfonts/, 18 MB
in total), as do the fonts and the rest of the interface: the app works offline and depends on
no third-party service. The official CDN is kept only as a safety net, in case a preset is
missing locally.
A note on the language. The code comments, the user manual and the commit messages are in Italian — it is the language this project is written in. The interface is being translated; until then, this README and CONTRIBUTING.md are your way in.
If suMidi is useful to you: ko-fi.com/johnwhale ☕
Released under the GNU Affero General Public License v3.0 — see LICENSE. In short: you are free to use, modify and distribute the code, including commercially, but if you distribute a modified version — including offering it as a web service — you must release your source code under the same licence.
A commercial licence is available. If you want to build suMidi into a product without releasing your own source, there is a paid exception to the AGPL: contact details are at www.johnwhale.com. See COMMERCIAL.md.
Pull requests are welcome — with one caveat: a change that alters the sound is a musical decision before a technical one, so open an issue first. How this project works, and what a change needs to be accepted: CONTRIBUTING.md.
For the commercial exception above to remain possible, every contribution goes through a short licence agreement — CLA.md: you keep your copyright, there is nothing to sign, one line in the pull request is enough.


