Polymodal Music SystemsInstruments for living music
A note to the Suno team, in response to your developer API invitation

Polymodal Music Systems · Seed Performances

A single idea, grown into a band you can perform against.

You asked to hear about work that would not be possible without your engine. This is that work. A Seed Performance takes one captured musical idea, sung, played, or spoken, and grows it into a living, performable backing in the idea's own key and tempo, which a person then performs against and records. It is already built and running. One piece, and only one, is beyond what any synthesizer can supply: a backing that sounds like a record. That piece is you.

If you read nothing else

Seed Performances are already implemented and running inside my CoDriver system. Capture, key and tempo detection, role muting, live performance, take management, and export all work today. The single limitation is that the backing is grown from MIDI and synthesis, so it sounds synthetic. Your generative stems, in the right key and tempo, replace that one weak link and turn a practice toy into a performance instrument. This is a use of your API that a browser cannot reach and that your web product does not capture today.

01What a Seed Performance is

A seed is the smallest capturable unit of a musical idea, and it can be any bounded creative act: a hummed melody, a two bar riff, a line sung into a phone. The premise is simple. A seed is valuable not for what it is, but for what it can open. A ten second idea is worth more as a doorway into a hundred performances than as one frozen file.

A Seed Performance is the environment that keeps that doorway open. It grows the seed into a full backing, hands you one role to play or sing live, and captures the take, while the seed itself persists, ready to grow again into a different key, a different arrangement, a different performance.

02Here is how they work

Four movements, from a fragment to a finished take. The colored strip below is the whole arc; the numbered steps are what actually happens.

Movement one
Plant

Capture the seed: sing it, play it, or speak it.

Movement two
Grow

A living backing is generated in the seed's key and tempo.

Movement three
Perform

Take one role and play it live against the band.

Movement four
Keep

Record the take. The seed remains, ready to regrow.

  1. 1Capture the seed. Sing a melody, play a riff, or speak a line. Anything with a start, an end, and a musical intent qualifies. Nothing has to be finished or perfect.
  2. 2Read the seed. The system detects the key and tempo, and, from an audio seed, the melodic contour and feel, so everything that grows next stays true to the original idea.
  3. 3Grow the backing. A full arrangement is generated in that exact key and tempo and delivered as separated stems, one per role: drums, bass, keys, guitars, and the rest.
  4. 4Choose your role. Mute the one part you intend to perform, the lead vocal, the guitar, the solo, and keep every other stem playing as a full band underneath you.
  5. 5Perform. Play or sing your role live against the living backing, at any blend from a silent part you own outright, through a soft guide, to a full reference you follow.
  6. 6Record and keep. The take is captured. The seed persists and can regrow into a new key, a new arrangement, or a different role on demand. One seed, many performances.
Already built. Every step above runs today inside my CoDriver system: the capture surface, the key and tempo reading, the per role stem muting, the blend control, live recording, take history, and export. The demos of the stem handling run in a plain browser with nothing uploaded. Only step three, growing a backing that sounds like a record rather than a synth patch, waits on a generative audio engine of your quality.

03Why the current version has a ceiling

The backing today is grown from MIDI and synthesis. The notes are correct; the sound is not believable. A triangle wave stands in for the lead, a bare sine for the bass, and there is no room, no breath, no ensemble. For private practice that is enough. But the moment a person wants to keep a take, share it, or build on it, the synthetic backing is the exact thing that makes the result feel like a toy instead of a record.

This is worth stating precisely, because it defines the whole opportunity. The limitation is not the note data and not the interaction design, both of which are solved and shipping. The limitation is that MIDI cannot produce a backing a person would be proud to perform over. Realism is the ceiling, and realism is a generative audio problem.

04Why Suno is the missing half

Every capability a Seed Performance needs maps onto something your engine already does, and does better than anyone. This is not a wish list. It is your current feature set, read as the missing half of an instrument that is otherwise finished.

Realism
Hyper-real audio, on key and tempo

The living backing stops sounding like a synth patch and starts sounding like a session. This alone lifts a Seed Performance from demo to instrument.

The exact mechanism
Generative Stems

Your engine returns a track already separated into time aligned stems. That is precisely what a performance environment needs: mute the one role the person plays, keep the rest as a full band. What I build by hand from exported stems, your API delivers natively.

The heart of it
Cover and audio input

The seed itself can be a performance. A voice memo or a played phrase becomes the seed while its melody and feel are preserved. Growing a work from a human performance rather than a text box is the core of this concept, and it is what your Cover pipeline was built to do.

Continuity
Personas, Voices, custom models

A backing and its re-voiced takes can carry one consistent identity across a whole body of work, turning a one off into a repeatable environment a creator returns to for months.

Where it happens. Through the developer API, all of this can run where the performer already works, inside their DAW, asked for out loud, mid session, with no website to visit and no download folder to manage.

05Why this cannot exist without you

Stated plainly: capture, key and tempo detection, role muting, live performance, take management, and export are already built and running. The single thing none of that can supply, and that only a generative audio engine of your quality can, is a believable living band grown from the seed on demand, in the right key and tempo, arriving already separated into stems.

Without your engine, Seed Performances stay a synth demo. With it, they become a performance instrument. That is exactly the class of project your developer invitation asked for.

06What is already built

  • Seed Performances V0 and V1, running today: plant, grow, perform, record, all working on MIDI inside the CoDriver system.
  • Three browser stem instruments (Stem Station, Stem Synth, Stem Sampler), built in a single day on your public output, turning stems into playable pads and keys with nothing uploaded.
  • A voice control bridge into a professional DAW, so generation is spoken for mid session rather than downloaded and imported.
  • A notation lane that turns generations into engraved, publishable sheet music, opening an education channel alongside performance.
  • A developer API application already on file, submitted July 4, with rights, licensing, and attribution handled with a practicing attorney's rigor, on verified sources only.

The ask

One design partner slot. An API endpoint and a month. In return: a real integration rather than a survey response, structured feedback on the API from someone shipping against it weekly, and a performance first use case, plus education and live, that your web product does not reach today. Give me the endpoint and I will bring you the demo you cannot build alone.

Start the conversationAlexander Scott Dennison · Polymodal Music Systems · polymodal@gmail.com