Wavefunction Collapse

I’m writing an album, Wavefunction Collapse: nine tracks of industrial / dark electro / goth house about a dead, uploaded mind orbiting the sun. It fuses a bunch of things I love, and mostly it’s an experiment to see how far AI-assisted music production has come.

It grew out of my Suno experiments. Once I saw how faithfully Suno could cover my own recordings, I started thinking about scaffolding a whole album from scratch and producing it with Suno plus other tools. So I decided to tell a story.

The story

It started with one track, now track 6: “Halting Problem,” about an uploaded mind and billions of its clones in a matrioshka brain parked close to the sun, remembering the flesh and wrestling with solipsism. It draws heavily on the Sobornost from the Quantum Thief books, some of Greg Egan’s work (Permutation City, Schild’s Ladder), and the series Pantheon.

I cloned my voice, wrote lyrics, used Claude to iterate on prosody and metre, and after a couple days had something I’m really pleased with:

Then I decided to turn it into an album. The premise: a human has died and been uploaded, and now runs as billions of connected instances aboard the Great Crystal Craft, a faceted vessel orbiting the sun and drinking its light to grow a matrioshka brain.

The Task is singular and total: collapse the star into a black hole and encode its wavefunction on the event horizon as a computational automaton that blooms into a pocket universe. A transhumanist riff on Fyodorov’s Great Common Task: defeat death by preserving, and ultimately re-creating, everything that ever was.

The catch: immortality through multiplicity looks a lot like ultimate self-absorption, and the album lives in that ambiguity. Nine tracks arc from cold certainty, through solipsistic vertigo and a grief-loop that refuses to halt, to a collapse that finally halts it and a genesis that is transcendence, but stranger than salvation.

The workflow

To write the rest I started using Claude Code as a co-writer, with a heavily modified version of the bitwize music skill. The whole album lives in a git repo, with Claude skills and hooks maintaining the authoring workflow: track files, motif and callback cross-references, style prompts, production notes.

My per-song loop looks like this:

Write out the story beats plus some lines I think would sound good. Have Claude generate several hundred candidate verses constrained to a vocabulary flavor, pick the handful I like, and scaffold them into a song.
Feed those lyrics into Suno with a basic style prompt to hear them.
Rework, regenerate, repeat until I’m happy with the words.
Generate several dozen versions with finalized lyrics and much more descriptive style prompts, until the vocals sound close to what I want.
“Cover” that song with my uploaded voice model. This gets somewhat close to sounding like me.
Record myself singing it the way the voice model does, use that recording as a new voice model, and cover the original again.
Now, to my ear, it sounds exactly like me. Or a version of me with a really good studio who warmed up perfectly and had a great session: what would normally take several days of takes.
From there, more covers with different style prompts. Suno is great at preserving vocals across permutations.

Partway in I realized I needed better raw material to feed Suno, so I built Motif Forge to generate the hooks and leitmotifs that enter this loop as seed audio.

Where things stood on July 9, 2026

A story I want to tell, fully mapped, with motifs and callbacks tracked in the repo.
Four tracks with final lyrics. Halting Problem (above) is fully produced and serves as the reference mix the others need to belong beside; the other three are below, and I like them but will keep iterating.
One track in progress (Census of One).
Four tracks still to write.
Motif Forge, built to assist in the hook creation process.
A set of Claude skills and hooks maintaining the album authoring workflow in git.

Here are the three I’ll keep iterating on:

Several missing pieces remain, mostly spot-editing parts of a song without losing the bits I like. Suno is very limited here, so I ordered a second 24 GB GPU to test local models that at least have API access. The local stack is still only a plan. Most of what frustrates me about Suno comes down to its web app and the manual work around it.

Some notes on all this

I had Claude clean up this post, because I wrote it chain-of-thought and tend to ramble. The thoughts are all mine. Same approach as the album, lol: I get bored once the work turns tedious (copy editing, grammar, coherence, hunting down links).

The creatives calling every use of these tools slop are missing what interests me: they let somebody work past the parts of a medium where limited technique used to stop the project cold. I am obviously approaching music like an engineer. I am also having a blast, which matters more to me than permission from the fun police.

Links

Album tracks so far:

See also: my Suno experiments and Motif Forge, the hook generator I built for this.