The Conflict between Generative Art and AI

Recently, I came to the realization that I needed a centralized archive of code based artwork that I’ve been creating with pen plotters since 2019. I meticulously dug up photos of my drawings, and documented their metadata in a single repository.

I came across a few photos of my earliest Generative Art algorithm, which I originally authored in 2019 in Processing.

It was spurred by the purchase of an Axidraw pen plotter, and I needed something interesting to put it to work with. The algorithm took a greyscale bitmap image and read its pixels, applying a random pattern to each based on its value. Soon, I figured out I could read each color channel individually and make my drawings in multiple colors.

Finding the above image of this old drawing brought a smile to my face, and I tweeted a picture of it.

A few days later, I checked X and my eye caught an uncanny image. It was my artwork, but in a different place, in a different size. It contained a near carbon copy of the forms and colors of my own creation, but it wasn’t anything that I had created. It was signed, in the bottom right, just like how I do it, but it wasn’t my signature.

Then, it happened again.

Browsing Instagram, I saw a Uuna Tek, a Chinese pen plotter company reposting work that looked eerily familiar. I’d already noticed that Uuna Tek plays fast and loose with what is “real” by using mostly AI images for their products, so they certainly don’t seem to care about misrepresenting the origins of specific artworks or motifs. And this isn’t a small-time company, they claim to have over 20,000 customers and have over 8,000 followers on Instagram alone.

The original artwork in question was my 2020 series entitled “A Series Of Tubes”. I stumbled upon various examples of my artwork, recreated rather poorly, and drawn with an Uuna Tek machine. Upon further investigation, the “creator” spun up a website to sell this “artwork”.

How was this possible?

Vibe coding, a lack of imagination, and a desire to make something cool.

It didn’t take very long to figure out that these were examples of people who had used an LLM to create their own version of my generative algorithm, and in remarkable time too.

In the past I’ve seen imitations of my artwork online, and it’s always been flattering and a bit awkward. I myself have been inspired by other artists and picked out specific techniques or ideas as seeds to my own work. This felt markedly different though. With no understanding or effort, the creations I held so closely could now be cloned with an astounding accuracy.

This sent me into a pure spiral. If anyone with a $20 LLM subscription can effectively build my portfolio themselves, what unique or competitive advantage (in a dire art market no less) do I have?

It became obvious that the more I shared about my art practice, techniques, and processes, the more often I could expect people to ask LLMs to replicate it.

This negative relationship between sharing ideas and automated theft is the antithesis of why I choose to create art. I’m motivated by using technology to connect with others, and I get a lot of satisfaction from freely sharing ideas and discoveries in public. Knowing that I am freely giving away my insights to enhance an LLM is not part of my charter, and yet, I am powerless to stop it. Unless of course I stop doing what I love.

Beyond destroying my online identity and fine art business, I briefly considered other options.

Perhaps I could outpace LLMs with more complexity in my generative codebases and exceed the depth of what vibe coders could throw tokens at.

As of late 2026, I believe my project BUSY can’t be agentically replicated. I spent a year on the code and illustrated 100+ little assets found through the 55 randomized drawings, which took around 750 hours to produce.

But if you’ve used frontier models in the last 12 months, you’ve seen the rapid curve of advancement, and must conclude that complexity is no true firewall. Furthermore, leaning into difficult problems as a basis for art is extremely limiting creatively and would be a nightmare to execute. Artists are not going to win an arms race against AI solely on the grounds of solving challenging computer graphics problems.

Art doesn't have to be hard to make to be good or important.

At one point I put out an open challenge to my followers on Instagram to see if anyone could vibe code an infinite polygon slicing algorithm that I wrote in an afternoon. The results actually surprised me with how close they got, although no one was able to fully replicate the example. If I were to put my code online, that door would be shut forever.

LLMs are creating major conflict within the Generative Art community

Generative artists, who now often refer to themselves as “Algorithmic Artists”, simply to create distance between themselves in Generative AI, are one of the many communities grappling with the role of AI in their ecosystem. We join the writers, the musicians, the designers, and many others as they contend with new social norms around defining the value of creative authorship.

Being adjacent to technologists broadly, generative artists have found their work being diminished into LLM benchmarks.

In a recent case, the artist Mannay spent a year writing the code for a new project, only to have the first public view of it vibecoded “in 20 minutes by Sonnet”, thus demonstrating that spending any real time doing anything yourself is clearly a waste of time.

Every time an artist opts to share their code on OpenProcessing, ShaderToy, or a public repo they’re unintentionally enriching the capabilities of these tools, and empowering those who seek to minimize and copy their creations. How does the community continue to be a creative, spirited, and inclusive contend with this new reality?

How do we continue to champion yearly events like Genuary when it is so blatantly obvious that they’ll be feeding grounds for LLM training, rip-offs, and low effort content.

The community struggles to answer these questions.

The continuum of AI Use in Generative art

So far, I’ve been identifying examples of people using LLMs to take content from X artist and autonomously replicate it into Y thing. This is not a creative pursuit, it is extracting an idea they didn’t have from online media and depositing it into a statistical recombobulation. Otherwise known as the vibe code.

I have probably led you to believe that generative artists are purists, who hand-author everything. This is sometimes the case, as my friend and fellow artist Piter Pasma insists on doing. But talking with other artists, there is a plethora of combinations of the artists ideas, code, and LLM usage.

I don’t think there is any right or wrong answer here, only the rules which I’m currently following.

1. Coding is art, and is an enjoyable experience that I do not want to retire from.
2. Therefore, art made by me is authored by me, and not agents. It’s slower, more error prone, and takes longer… and that’s just how I like it.
3. LLMs are great serving as super-Google, and I use them to understand new computer science concepts.
4. Vibe coding utilities and applications which are paywalled feels okay. I derive no pleasure from setting up services in Linux to do mundane tasks, and will happily delegate some of that work.

I want to acknowledge my privilege in being able to set these boundaries. Most everyone in software engineering does not have this luxury. The craft of programming may very well be left to those of us artisans who build the weird things.