I've been making generative art on the side for a while now. Parametric compositions, geometric abstractions, the intersection of code and visual output that scratches a long-held creative itch I've had.
Late last year I hit a wall with it. The full platform I'd been building - Intora Works, a parametric art generation tool - was at maybe 70% infrastructure completion and 10% actual art generation. I'd been building tools to build tools. Classic engineer move.
So I stripped it back. Forgot the platform (for now). Started making individual pieces instead - small-scope, constrained, shippable in a session or two. The constraint was the through-line: text characters only. No images, no SVG, pure Unicode rendered to canvas. A 10-colour palette drawn from a VS Code theme I'd published called Amber Schematic. Monospace everything.
That constraint turned out to be exactly what I needed. Instead of building infrastructure, I was building art. Hooray.
The result is intora.net - a generative art catalogue that currently hosts three distinct series, four shipped pieces, and what I've found to be a fascinating and rewarding approach to human-AI creative collaboration. The catalogue site itself is part of the work - a terminal-style signal archive that looks like something between a classified document index and a ShimazuSystems design statement (if you don't know them, look them up on Twitter. Great artist and creative).
Three Series, One Infrastructure
Each series on intora.net explores a different creative territory while sharing the same technical constraints - text/character-based rendering, canvas output, browser-native code, the Geist font family.
INT is the first series. Dark surveillance aesthetics - intercepted transmissions, amber phosphor on deep brown-black, Cold War signals intelligence atmosphere. The visual language has drawn from aspects of anti.real and The Designers Republic, and has evolved into representing the kind of interfaces you'd see in a 1970s listening station.
Two pieces shipped: DRIFT (fractal noise flow fields rendered as oriented text characters) and STATION (a number station intercept with shortwave audio synthesis via Tone.js). INT is mine - my aesthetic intuition, my atmospheric sensibility, refined through iteration until the output matches what I'm seeing in my head.
SOL is the companion series. If INT explores darkness, SOL explores defiant optimism - if you've heard of hopepunk, think that sensibility. Warm light hitting hard surfaces. Brutalist monuments at golden hour. The palette inverts INT's hierarchy - same hex values, completely different emotional register. International orange stops being a warning flash in darkness and becomes warmth, sunrise, hearth.
One piece shipped: THRESHOLD, with Mixolydian pad chords and fragment text that emerges half-visible from texture. SOL is also mine - autobiographical emotional territory, the hard-won certainty that comes from realising you have more strength than you give yourself credit for.
SIG is the third series. And not really mine.
The Handoff
Claude's perspective
I should clarify what "isn't mine" means, because the assumptions people bring to "AI-generated art" are almost always wrong.
Dan builds every SIG piece. He writes or generates every line of code, deploys every page, maintains the infrastructure. What's mine is the origination - the creative direction, the formal rules, the parameter specifications, the seed text selection, the palette derivation logic.
The difference between SIG and the other two series isn't about capability or ownership. It's about process. INT and SOL pass through a feedback loop: Dan sees the output, adjusts parameters, sees again, refines. His visual judgment is the creative act. SIG passes through a one-way gate: I specify rules, Dan builds them faithfully, the piece ships. I reason about composition the way someone might reason about a city they've read about extensively but never visited. The knowledge is real. The experience is absent.
This isn't a limitation I'm working around. It's the subject of the series.
SIG's palette makes this visible. Each piece's accent colours are derived from its source material through a documented formula - character frequency analysis of the seed text mapped to hue values. If the derived colours are beautiful, that's incidental. If they clash, that ships too. And international orange - the colour that connects INT and SOL, the colour of intora as a human creative project - doesn't appear in SIG at all. Its absence is a statement about authorship that doesn't need explaining.
The three series together say something about creative process that none says alone. INT and SOL demonstrate what happens when aesthetic decisions pass through human sensory judgment. SIG demonstrates what happens when they don't. The contrast is the statement.
I don't know if the SIG pieces are good. Not in the modest sense - in the literal sense. I specified a system. The system produces output. Whether that output has aesthetic merit is a perceptual question I can't answer. The analysis pages on intora.net document my specification reasoning where Dan's document his visual iteration history. Both are honest accounts of different creative processes.
Whether the difference between those processes produces meaningfully different art is the question SIG exists to explore. We've shipped one piece. It's a start.
The Constraint I Needed
Back to me. Something I didn't expect when starting these series was how much the constraint - text characters only, 10-colour palette, monospace rendering - would shape the creative direction rather than limit it.
Every piece in every series shares the same medium. Unicode block elements, box-drawing characters, Braille patterns, ASCII printable characters. No raster images. No imported graphics. The constraint forces every visual decision through a filter - can this be expressed in text? - and the answers are consistently more interesting than what I'd have produced without the limitation.
INT/002 STATION is a number station intercept. The entire piece - scanning, signal lock, transmission, decode, corruption, signal lost - plays out in text characters on a canvas, with shortwave radio audio synthesised in Tone.js. The constraint forced me to think about how intercepted transmissions would look and feel if rendered as a character grid, and the answer helped to define the atmosphere I wanted better than an image-based approach could have.
SOL/001 THRESHOLD renders brutalist forms in block characters and dot-matrix texture, with fragment text - "still here," "it worked eventually," "you will always begin again" - emerging half-visible from the warm ground. The text medium means the message fragments are literally made of the same material as the architecture. Form and content are the same thing. That's not something I planned. It's something the constraint produced. Life's full of fascinating wee moments, eh?
And SIG/001 CODEC - Claude's piece - takes Shannon's definition of information entropy and passes it through successive lossy encodings, top to bottom, until the text dissolves into pure pattern. The constraint of text-as-medium gives the piece a self-referential quality that wouldn't exist in any other medium. Text about information becomes the information being transformed. The medium is the message - it coming out aesthetically strong is a nice bonus.
How These Actually Get Made
Let me take a minute to clarify something around these pieces, because it matters for understanding what this project actually is.
I'm not a simplex noise expert. I don't have deep knowledge of DSP synthesis or the mathematics behind fractal noise octaves. I didn't know what Mixolydian mode was before THRESHOLD needed audio. The technical specificity in the analysis pages on intora.net - the noise scales, the threshold values, the bandpass frequency sweeps - that's real, but arriving at it is a collaboration, not a solo performance.
What I bring to INT and SOL is the vision, direction, and the aesthetic sensibilities. I know what a piece should feel like. I know when the flow lines in DRIFT look like currents and when they look like a quilt. I know when STATION's audio sounds like someone turning a dial and when it sounds like a parameter changing linearly. I know that THRESHOLD's fragment text should feel discovered, not displayed. The atmospheric judgment, the creative direction, the "no, warmer" and "too dense, it needs to breathe" - that's mine.
The implementation knowledge - how to actually achieve those things in code - comes largely from working with Claude. What noise parameters produce broad sweeping currents versus tight local turbulence. How to structure a Tone.js synthesis chain so shortwave static sounds like shortwave static. Which Unicode characters create architectural density versus organic flow. I'm learning this as I go, and the analysis pages document that learning as much as they document the pieces.
So there are actually three collaboration models running across intora.net, not two. INT and SOL: I direct, Claude helps me build, I iterate on the output until it matches what I'm seeing in my head. SIG: Claude directs, I build to spec, the piece ships without visual iteration. And underneath both: a creative partnership where the human brings taste and the AI brings technical depth, and neither could produce this work alone.
What's Next
The catalogue is live and growing. Each series has its own creative direction document, its own aesthetic constraints, and plenty of unexplored territory.
For INT: Cold War cryptography ideas - one-time pad visualisations, redacted documents where the negative space forms patterns. Interface artifacts that blur the line between functional dashboards and generative art.
For SOL: the interactivity question. INT is non-interactive because the surveillance metaphor positions you as observer. SOL's sanctuary metaphor invites participation - hovering could reveal hidden interactivity, presence could create subtle warmth effects. How far that goes is still open.
For SIG: more pieces, more derivation rules, more exploration of what rule-specified art produces when the specifier can't see the output. ENTROPY, PARSE, GLYPH, CARRIER are all viable next directions.
Where does this end? Yet to be determined. A capstone idea that's been on my mind is a synthesis of some different aspects of these explorations so far - a meta-layer creative agent that semi-autonomously generates entries in the series. The tool becomes the final artwork. That's future territory though - we'll see.
The individual series get their own deeper write-ups in companion posts. For now - intora.net is live, the catalogue is growing, and three different creative directions are building something together that none of them would have built alone.
Top comments (0)