Jun 20
BGAN Punks: Transformative. Not Derivative.
Bastard GAN Punks V2 — or more simply BGAN Punks — are a collection of 11,305 NFTs on the Ethereum blockchain. They were first minted in March 2021 and each BGAN shares the DNA of their CryptoPunk parents even though their Punk parents are unknown. BGAN Punks V2 predate nearly every other successful NFT collection including Bored Ape Yacht Club, Cool Cats, and even Art Blocks. 847 are animated gifs, and 2,054 BGANs integrate a second project called GLICPIXXX into their background image. BGAN Punks V2 are the most successful AI-generated NFT collection to date. See the entire collection here.
For BGANPUNKS V2, I chopped all CryptoPunks attributes, made cross-gender copies of ones which were made for only one gender (BGANPUNKS ARE QUEER!), categorised them in different folders (Hairs/Heads, Eyes, Glasses, Smokes etc.), and wrote a script to make all possible combinations of those traits. I also added skin and accessory color randomization + a couple of custom traits I’ve drawn myself. To be honest it was a painful and shitty process because I had to do all the cuts by hand and had to optimize the script via trial and error maybe like hundreds of times, to not to make [an] unnecessary 1,000,000 images at the end which would take a lifetime to train the model with my shitty GPU. — Berk Özdemir
CryptoPunks were pieced together through a generated code. Meaning, someone at Larva Labs drew out each trait, and plugged them into an algorithm in order to create the 10,000 unique images in the NFT collection. Many many other generative collections were made this way; though, not all generative NFT collections are in a profile picture format. Art Blocks gave generative artists who work with code a platform with a high standard of quality to present their art to the world.
“Art Blocks is a first of its kind platform focused on genuinely programmable on demand generative content that is stored immutably on the Ethereum Blockchain. You pick a style that you like, pay for the work, and a randomly generated version of the content is created by an algorithm and sent to your Ethereum account. The resulting piece might be a static image, 3D model, or an interactive experience. Each output is different and there are endless possibilities for the types of content that can be created on the platform.” — artblocks.io
CryptoPunks and Art Blocks share their generative genes, and both are wildly successful in the web3/NFT space in terms of recognition and monetary value, but people collect them for different reasons. With Art Blocks, collectors are drawn to the shapes, depth, patterns and nuances the artist’s code has created. In their Twitter thread Bonafidehan outlines what makes Tyler Hobb’s Fidenza 938 so desirable. They define its name God Mode, the algorithmic traits at play, and the illusion that work in unison to create the piece. But also note that saying the algorithm did all the work is just flat out incorrect. Tyler Hobbs knew the possible outcomes when he wrote the code to create the collection. Though, probably just not the exact outcomes the code would produce.
But for CryptoPunks, it’s a mater of identity. In 2017 CryptoPunks were created and given away for free to anyone with an Ethereum wallet. The blockchain was just two years old, and Larva Labs created an online face for the growing user base on social media to signal each other. This social signaling allowed like-minded people to talk about about Ethereum in a safe, anonymous way. This was no doubt very important in the bear market of 2018-2021. And so CryptoPunks earned their prestige as they helped shape the norms we have today across all of web3. This is why people perceive CryptoPunks as crypto OG’s and new-comers often listen to what they have to say.
But as time goes on we see the legitimacy of CryptoPunks start to falter as people who can afford them start buying their associated prestige. And so as the space continues to grow and more nuanced sectors begin to form, the words of CryptoPunks begin to carry less weight.
GAN VS. The Human Hand
CrypoPunks’ traits were all hand drawn and then pieced together randomly by code. BGAN Punks’ were not and this is where the biggest difference lies. Since BGANs are human faces imagined by a machine, the outcomes were unknown and uncontrolled by the artist. The GLICHY AF trait is the only exception to this. To achieve this trait Berk used “programmatically tiled images that are encoded with open source software called GLIC.” Which they integrated into the V2 collection. And is also the project known today as GLICPIXXX.
Generative Adversarial Networks, or GANs for short, are an approach to generative modeling using deep learning methods, such as convolutional neural networks.
Generative modeling is an unsupervised learning task in machine learning that involves automatically discovering and learning the regularities or patterns in input data in such a way that the model can be used to generate or output new examples that plausibly could have been drawn from the original dataset.
GANs are a clever way of training a generative model by framing the problem as a supervised learning problem with two sub-models: the generator model that we train to generate new examples, and the discriminator model that tries to classify examples as either real (from the domain) or fake (generated). The two models are trained together in a zero-sum game, adversarial, until the discriminator model is fooled about half the time, meaning the generator model is generating plausible examples. — MachineLearningMastery.com
The only resemblance to CryptoPunks is the 24 pixel square in which the Bastard takes its form. BGAN 619 is even unlike any other BGAN Punk. its Double Pipe ‘n Big Smoke trait is the only occurrence in the entire collection. With the combination of its other traits: Front Facing, X Bastard, and the Eyes background pattern make 619 a Grail Bastard. When considering the aesthetic appeal of BGAN Punks, think about the ensuing battle of the GAN that lead to the Bastard’s creation. The generator model follows Berk’s guidelines in order to defeat the discriminator model’s idea of how the output should look. No doubt outputs like 619 are part of the learning process, but the pixilated noise is what captivates the eye. This is the entropy of us. As the world naturally trends towards chaos, so too does the machine’s perception of humanity, as this noise is persistent throughout the collection.
Also note that none of these traits are present in the CryptoPunks collection. Pipes are, but only in the bottom right corner and only colored brown. But the generator here makes its own rules placing a pipe in each corner, maybe fading one just enough to fool its adversary, but firmly deciding on green. The generator is fixated on the left pipe. It is perfectly defined. Surly this alone will thwart the discriminator.
The X Bastard’s pareidolia face is a trait found in BGANs created early in the machine learning process. The aesthetic appeal comes from the human eye’s desire to find human faces is unsuspecting places, like in a flower or the windows of a house. The brain is not only wired to see faces in inanimate objects, but also to label these faces with emotions. In BGAN 619 there is a crooked smile of light pixels, or, no, perhaps the lips are pursed as he draws from the pipe. The eyes are straight on and piercing. Is 619 listening intently? Does the pipe’s fade from right to left suggest disagreement? And the ghost of a Big Smoke maybe says, “this conversation has gone on long enough.”
The Glitchy background shows a flaring eclipse on purple sky highlighted in white with a contrasting green bar breaking across the full length of the top of the portrait. These backgrounds though not created by GAN, were generated through code controlled by Berk. And then paired with a complementing BGAN. This extra attention sets BGAN Punks in a league of their own. But with the very unique qualities of 619, one has to consider the the intentional decision in the artist’s pairing. There is a mood and depth here. There is a message of doom meant to be seen and heard.
This effect is found in most BGAN Punks. The observer is able to find relatable nuances in the face and accessories, with which they identify. Or evokes emotion through the colors, patterns, and imperfections found in the square pixels. This along with the same identifying features found in the CrypoPunks collection, where one identifies with a Punk because it looks like them. It’s no wonder collectors are so passionate about their BGANs when each owner sees themself in their Bastard.
Bastards are the Future
With the current market conditions web3 communities have started to see a great exodus in participants. But as conditions improve these people will return, and more adopters will come with them. And so surviving these hard times is paramount to succeeding in the next cycle. As time moves on and new arrivals come seeking information, it will be the BGAN Punks who survived sharing it with them. It will be these survivors inviting them into the Bastard DAO. And in turn these new Bastards will learn how to survive hard times.
It is our responsibility not only to survive, but to learn so we may teach and to build so we may share. The timing of BGAN Punks’ creation contributes to the collection’s legitimacy, but our community’s willingness to contribute to the space as a whole along with our culture of inclusivity is what will lead us to success. In the coming months as community members continue to legitimize the collection through careful brand identities of podcast hosts, critical thinkers, builders, and esteemed collectors we will start to see a shift in public perception of BGAN Punks as the collection in turn begins to legitimize its owners. And in the near future we may very well see BGAN Punks showcased in prestigious collections alongside CryptoPunks and Art Blocks, for their representation of NFT culture, technology, and art.
