31 Oct 2023/40 min/Pindar Van Arman
MLow X Pindar Van Arman
Twenty years of painting robots, a 100 percent on chain collection dropped the day OpenSea killed royalties, and why beautiful AI makes story the scarce good.
Pindar Van Arman has spent nearly twenty years, since 2005, programming painting robots with AI, and this conversation catches him mid transformation: from an AI artist who used the blockchain to sell art to a crypto artist who uses the blockchain to make it. His ByteGANs, dancing pixel aliens dripping neon slime, are 100 percent on chain, every bit inscribed on Ethereum with no IPFS and no external dependencies. By pure luck they dropped the day OpenSea abandoned royalties, and he watched royalty payments collapse to near zero within days. He had the option of a royalty guard or operator filter and refused it on principle: encumbering the contract would violate the piece's own ethos, even at his own cost. Half his collectors treat the work as art, half trade it like an altcoin, farming Blur with it toward roughly 6,000 ETH of volume, and he accepts that an artist who releases work into the wild loses control of it.
The robot segment is the fullest technical account in the archive. His current system, feedback, paints with expression, draws with technical precision, and uses reflection to decide which comes next: expressive brushstrokes from generative neural networks, technical marks from a second head holding a pen, a photograph of the canvas every hundred strokes, one to three days and thousands of decisions per painting. He cites Marvin Minsky's Society of Mind, creativity as a thousand competing processes, admits he failed two neural network courses before page 127 of Grokking Deep Learning delivered his eureka, and trains models only on his own pixel art and a lifetime of his own portraits, refusing to prompt in the style of any other artist.
The forward looking core is his prediction that as AI makes all imagery beautiful, colors perfect by default, the story behind the work becomes the differentiator, unleashing people who always had a vision but lacked technical skill. Around it: plotter advice over robot arms, ChatGPT compressing three days of coding into about an hour, the Sovereign drop studio for indie artists, regret over his old BitGANs raffle gimmicks, a no gimmicks no utility ethic he keeps despite collectors dumping over it, and a promise to give every past collection attention for the rest of time.
What this episode teaches
- The medium is the statement
A bronze reads differently from a plastic sculpture, and a piece inscribed 100 percent on Ethereum says something a hosted JPEG cannot. Van Arman built ByteGANs so they dance forever, as long as Ethereum exists, and treats that permanence as part of the art itself.
- Refuse enforcement that betrays the work
He declined a royalty guard and operator filter on ByteGANs because a pure, unencumbered contract matched the spirit of the piece, even though the drop landed the day OpenSea abandoned royalties and his royalties fell to near zero. In hindsight he still calls it the right choice.
- Beauty is commoditized; story is scarce
Early AI was grotesque, monster faces at 128 pixels; now diffusion gets the colors perfect automatically, the hardest part of painting. His conclusion: when all imagery is beautiful, the story must be that much more compelling, and people with vision but no technical skill are about to produce an explosion of art.
- Train only on your own corpus
He is uneasy with prompting a sunset in the style of Picasso and will not risk exploiting artists or photographers whose work fed a model. Every ByteGAN trained on pixel art he generated himself; his face models trained on a lifetime of his own portraits. An ethical moat that is also a stylistic one.
- Buy a plotter, not a robot arm
For artists wanting robotic painting, Processing has plotter and XY table support, and a plotter is cheaper and more precise than a robot arm. The arm is performative, buying perceived agency and emotional effect, not better output.
- ChatGPT ended the software engineer requirement
Converting ByteGANs into plottable SVGs was budgeted at three days; with ChatGPT writing the program it took about an hour. His advice to artists experimenting with AI: anything you do, ask ChatGPT, in plain English, even to write AI itself.
- Effort is the recipe
Asked for the recipe for great art, he gives one ingredient: the single trait all awesome projects share is high effort. Spend four months on a project and he is not sure how it could be bad; 500 pieces each given individual attention is how his serials earn fans.
- Gimmicks work short term and poison long term
He regrets his old raffles, retweet for a waitlist spot promotions and bidder editions, which turned art into fomo and gambling; one underbidder confessed a 12 to 14 ETH bid was his life savings. He now runs no reserve auctions and refuses utility promises, even watching a collector dump ByteGANs over it, because incentives always end in disaster.
- Serve every collection forever
His legacy rule: every project he has ever made gets his attention for the rest of time. He buys his own ByteGANs back, has his robots paint physicals of them, pairs them with SuperRare purchases, and keeps telling their stories years after mint, royalty income or not.
The conversation, topic by topic
ByteGANs: 100 percent on chain
His latest mint, a neon dripping alien from the ByteGANs series, is inscribed entirely on Ethereum: no IPFS, no code dependencies, nothing external. He compressed 10 megabyte AI animations to 1 kilobyte with a custom algorithm so they fit on chain and still look cool, and calls the series his transformation from using the blockchain to sell art to using it to make art.
The dancing coin problem
Collectors split ByteGANs in half: art on one side, altcoin on the other, with Blur farming driving roughly 6,000 ETH in volume and a 30 ETH top sale. Dropped by pure luck the day OpenSea abandoned royalties, the collection watched royalties decay to near zero within days. He predicts phases: coin traders exit, hardcore holders accumulate, and the work is finally read as art.
From imposter to crypto artist, physical kept separate
His first SuperRare drops were photos of robot paintings, and he felt like an imposter next to natively digital artists like XCopy, Hackatao and Coldie, making AI art and merely selling it on crypto; five years later the work is pure code on Ethereum. Yet he never pairs physical and digital: a digital exploits impossible neon greens and motion, the robot painted physical has infinite resolution, texture, drips and visible mistakes, and buyers of his digitals may take the physical only as a second, distinct artwork.
How feedback paints
His newest robots, both named feedback (the name belongs to the system, more than 12 and fewer than 24 robots into the practice), paint with expression, draw with technical precision, and use reflection to choose between them. Every hundred strokes they photograph the canvas, check faces, emotion and contrast, sometimes scrap the face and reroll, and finish in one to three days over thousands of decisions.
Society of Mind, not one network
He rejects the idea that creativity is a single process, citing Marvin Minsky: a thousand creative processes compete and the situation picks the winner, exactly like his robot choosing between expressive and technical algorithms. He also recalls gatekeepers who once said only GAN art counted as AI art, a claim diffusion erased.
Learning AI by force of will
Sparked when GANs got hot and AlphaGo beat Lee Sedol, he took two neural network courses and failed both, defeated by linear algebra and calculus, until Grokking Deep Learning gave him a page 127 eureka. He calls himself book learned and no expert, integrating better neural networks from GitLab into a system that is nonetheless, he believes, the most creative AI art system anywhere.
Advice: plotters and ChatGPT
For robot curious artists: Processing supports plotters and XY tables, which beat expensive performative robot arms on precision. For AI curious artists: ChatGPT is superpowers; it turned a three day SVG conversion job into about an hour, and he jokes government software engineers keep a gentleman's agreement never to tell management how good it is.
From grotesque to beautiful AI
His early robots redeemed ugly 128 pixel GAN faces by painting them at infinite resolution. Discussing Robbie Barrat's early portraits, he admits he printed one, hung it, and took it down because it kept reading as Hiroshima burnt bodies. Now AI nails perfect color relationships automatically, so prettiness is table stakes and story becomes the differentiator; he dreams of AI helping him finally compose music and write a novel, and half hopes the AI industrial revolution brings two day work weeks.
Die With The Most Likes and Marfa
He saw DWTML's AI collection (roughly transcribed as Beef Rothko) the night before launch while in Marfa: 666 pieces, a flea market parody of the auction houses, built by training AI on a smaller set of hand painted canvases. For Van Arman it proves the thesis: vision plus story, amplified by AI, produced what six hand painted canvases never could.
Sovereign, the indie drop studio
With two partners he runs Sovereign, an indie label equivalent for sovereign artists: independent drops, no backing, no passes, every drop unique. He advises artists on making 500 piece serials work, an Art Blocks style mechanism for non generative artists who hand make each piece; recent sold out drops include a 500 piece collection and a hand drawn 500 piece set, and a MLow drop is in the works with his partner Ezra.
Favorites and the gallery walk
Recent buys: David Henry Nobody Jr (in love with the madness of his drops) and the DWTML AI drop; he limits collecting to portraits to avoid going broke. Walking his CryptoVoxels gallery live, he names Hackatao, Bard Ionson, Max Osiris, Angie Taylor, Gary Cartlidge and Patrick Amadon among five years of collecting, and calls Nori Harmon (roughly transcribed) and Die With The Most Likes his picks to break from crypto into the traditional art world.
Pricing, serials, and no reserve
His auctions run no reserve, a risk few artists take. Having learned 1/1s have a limited shelf life and collector type, he makes 500 piece serials at around 0.1 ETH, never editions and never an open edition, every piece individually attended to, trading margin for far more collectors and fans.
The story of the work, and Decoherence's puzzle
His singular tale: twenty years struggling to sell robot paintings above 500 or a thousand dollars, then crypto art, then the realization the medium matters as much as the art. Underneath sits his core question: the robot is unquestionably creative, so does it have free will, and do we? He drops an exclusive clue for the show: Decoherence hides an unsolved puzzle, 11 glitches plus a random word forming a seed phrase to a wallet holding a rare BitGAN, and his Discord comments are clues in disguise.
Quantum portraits and the Impressionist dream
His Quantum Faces portraits use quantum computing randomness to render subjects in two places at once; he trades them rather than sells them, just finished the first hundred, and books MLow for one on the spot. He closes with his college daydream of standing with the ridiculed Impressionists: crypto art is getting the salon treatment Gauguin, Toulouse-Lautrec, Van Gogh and Monet got, and this movement is international, so he expects it to be bigger than Impressionism.
In their words
“When you make a digital piece and you put it 100 percent on chain, that's making a very, very specific statement about that piece of art.”
Pindar Van Arman“Feedback paints with expression and technique, guided by reflection.”
Pindar Van Arman“The question that's core to my art is, what is creativity? What am I as an artist? How much free will do we have when we create?”
Pindar Van Arman“Now all the imagery coming out of the world is that much more beautiful. It means that the story has to be that much more compelling.”
Pindar Van Arman“The number one thing that all awesome projects share is high effort. If you spend four months on a project, I'm not sure how it could be bad.”
Pindar Van Arman“Not like you're just going to DALL-E or Midjourney and typing in a few prompts. You've built that framework, and that's why you're able to get to that output.”
MLowCite this conversation
Pindar Van Arman in conversation with MLow, "MLow X Pindar Van Arman," The MLow Show, 31 Oct 2023. https://show.mlow.xyz/e/2023-10-31-mlow-x-pindar-van-arman.html. Quoted from a machine transcript.
Quotations come from machine transcripts. Check wording against the audio before republishing.