Author: Mason Marcobello

The History of BGAN Punks and a discussion with their Creator

Posted on by Mason Marcobello

“REKT DEKT DEKT, GELD I’VE BEEN WALKING HOME IN SONA, I’M NOT GOING HOME I’VE BEEN WALKING HOME, THERE’S NO MONEY.” 

Such are the indelible lyrics of Mark Cuban’s Bastard. More specifically, one of his 12 BGANs (short for Bastard GAN Punks). Created by Berk Özdemir, the V1 and V2 collections are renowned for being fashioned by a specialized branch of AI known as Generative Adversarial Networks (hence the acronym “BGAN”). 

(One of Mark Cuban’s V2 BGANs)

With notable holders like Cobie and Punk 6529 (not to mention recently gaining a place in FloorDAO’s Treasury), BGANs are a homage to – and evolution of – the original CryptoPunks by Larva Labs. Overflowing with a layered creative ecosystem that predates Art Blocks, some consider them to be true digital art. 

The V1 collection of 76 NFTs was released in 2020, followed by the much larger V2 BGANs (11305 in total), minted on March 7, 2021. This was one month earlier than the much talked-about Bored Ape Yacht Club art NFTs. Both series include a range of genderless, carefully curated traits (unlike CryptoPunks). And, as mentioned, the core stylistic influence and aesthetics are determined entirely by AI. 

(BASTARD GAN PUNKS V1)

What are Generative Adversarial Networks (GAN)? 

A generative adversarial network (GAN) is a machine-learning model introduced by Ian Goodfellow in 2014, whereby two neural networks compete to create more accurate, or better fitting, data. They are used to generate new data for scenarios across sectors. Ian Goodfellow’s appearance on the Lex Fridman Podcast explains more.

Use cases are also found in contemporary art: content like images, audio, video etc. can be given to the neural networks to learn from, as the two models compete with each other to output higher-scoring answers. After a training session, new file(s) are generated that contain the AI’s interpretation of the original dataset and what it learned. 

Berk’s integration of GAN technology was a case of creative experimentation while undergoing a Master’s programme at ArtScience Interfaculty. 

Unable to afford an original CryptoPunk, Berk decided to make “interesting enough” remixes of them. BGANs remix “genes” representing a variety of traits from several original CryptoPunks, so there is no direct equivalent or correlation between them and the original NFTs by Larva Labs. Nobody knows which specific Punks were used to create the new interpretations by Berk, thus the origin of the name and concept “bastards.”

BASTARD GAN PUNK V2 #11188

When detailing his creative process, Berk states: 

“The first iteration of Bastard Gan Punks (v1) were trained on 10,000 original Cryptopunks and gave birth to many weirdo adorable motherfuckin’ bastards. The process worked very well and I was so happy with getting weird interesting results. However, since some traits of Cryptopunks exist on just a few bastards (like AliMason Marcobello talks to the creator of BGAN Punks, a collection of algorithmically-generated NFT artworks, about art, technology, and creation.en, Ape, Zombie, or Beanie etc), most of the rare traits were lost during the model training.

For BGANPUNKS V2, I chopped all Cryptopunks’ attributes, made cross-gender copies of ones made for only one gender (BGANPUNKs are queer!) categorized 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 hundreds of times and not make a million unnecessary images at the end, which would take a lifetime to train with my shitty GPU.”

Separate smart contracts for IP and licenses (commercial rights):

Alongside AI, BGANs have also catered to the obfuscated and at times contentious nature of intellectual property by granting full commercial rights to holders. Through a separate smart contract, holders have the option to tailor specific conditions of the IP for each NFT directly on the Ethereum blockchain.  

Traits and the BGAN Ecosystem:

Unlike the static metrics that help collectors filter through an ever-expanding constellation of NFTs, BGANs were released mostly without any pre-determined trait categories except for CALM AF, HYPE AF, and ‘Bastardness’ (a broader category with varying rarity metrics). 

Instead, over the ensuing months and through a meritocratic process aligned with the ethos of Web3, the community decided on the specific traits they wanted BGANs to embody. This initiative was later streamlined through a living document (authored by Snooplion) known as the Trait Constitution:  In essence, if any holder believes that BGANs should change or modify their traits, all they need to do is follow the process of the trait constitution and submit a snapshot proposal for community voting.

The current definitive traits of BGANs include:

  • HYPE TYPE (animated/still). There are 847 animated or “HYPED AF” vs. 10,458 still or “CALM AF” BGANS.  
  • SONG WORD COUNT
  • BASTARDNESS
  • BACKGROUND (still BGANS)
  • GLITCH PATTERN (still BGANS)
  • BAD HABITS (still BGANS)
  • FACING DIRECTION (still BGANS)
  • TYPE (still BGANS)
  • BACKGROUND GLITCH LEVEL (animated BGANS)
  • BACKGROUND MOOD (animated BGANS)
  • FLOATY HEAD (animated BGANS)
  • SPEEDOMETER (animated BGANS)
  • NUM OF FRAMES (animated BGANS)
  • HEAD TURNS (animated BGANS)

A more detailed explainer on how to sort through and understand these traits has been written by 888mooncat.

Adding to the world of BGANs, Berk also created the Glicpixxx (V1 & V2) collections, Artglixxx collection, and an associated $Glix Utility token.

Glicpixxx V1 

Glicpixxx V1 is the first of the Glicpixxx experiments by Berk. As with the lighthearted origins of BGANs, the idea of Glicpixxx started in 2020 after a streaming session on Twitch when Berk opened a favored encoder software (a processing program) called “GLIC” and started playing around with images. After taking one of the encoded images, he divided it into 36 individual squares on Photoshop and minted them (as a test) using OpenSea’s storefront contracts. 

GLICPIXVER001 CANVAS – #0

However, before releasing the GLICPIXXXVER002 GRAND COLLECTION, Berk updated the parameters for the original smaller series (V1) so that holders of the originals (that were created using the OpenSea storefront) could transfer it to a new contract and mint an equivalent copy. This was largely because (at the time) OpenSea used unverified contracts and stored all data on their API. Not to mention that the tokens functioned as singular ERC-1155 rather than the conventional ERC-721 standard for NFTs.

A unique aspect of the new contract for GLICPIXXXVER001 that Berk created with his friend (CANER) is the added NFT (TOKEN ID #00) .It functions as a programmable canvas that reflects all 36 GLICPIXXXVER001 in 6×6 tiles. 

Similar to async.art (or related programmable blockchain projects), the separate 36 GLICPIXs function as layers for the master #00 token. Any holder of the original series can change the render of the master token by flipping their NFT, rotating it, or swapping its position in the canvas with someone else. 

Each time a token holder changes the parameter of the layer they own, they also change the state of the (master) contract, and as Berk stated, “a P5 sketch rendering the tile canvas will start reading from these new parameters.” The code can be found inside the smart contract. 

Glixpixxx V2 – Grand Collection

On the same day he experimented with GLICPIXXXVER001, Berk generated an additional 48 images (1024 x 1024) with GLIC – all with a different range of presets, codecs, and parameters. A Python script was then used to explore the visuals in more detail, resulting in an additional 40k (32 x 32) source images, each with a unique and complex variety of patterns and colors. 

GLICPIXXXVER002 #5770

While preparing for the launch of BGANs V2 in 2021, Berk used a random selection of the GLICPIXXX as backgrounds for transparent Bastards and found an organic synergy with the aesthetics of the two collections. It wasn’t before long that positive acclaim spread throughout the BGAN community, with people hunting GLICPIXXX, the most striking combinations. 

Although the 40k source images were originally going to be combined into a series of artistic GIFs for VIV3, after a lull in communication, Berk decided to release it as a grand collection and gift for Bastard Gan Punk holders. 

As such, GLICPIXXXVER002 was born as an ERC-721 collection consisting of 11,500 images of 32 x 32 pixels. As with V1, V2s are programmatically encoded with open source software called GLIC. The collection consists of several subcategories and unique traits that reflect the native lore of BGANs. More details can be found here. 

Mirroring the creative process for BGAN Punks, GLICPIXXXVER002 images were created without the predefined traits or patterns usually associated with the “layer merging, generative method.” Instead, Berk added traits according to his analysis of each separate image, and the patterns were reverse-engineered, categorized, and assigned to BOOMER CALM AF GLICPIXXX by the BGAN Punk community. https://glicpixxx.love/readme 

As an ERC-721 token with an open license, GLICPIXXX can be used by anyone as a raw asset for creative initiatives or projects like virtual worlds, games, smart watches, fashion, texture packs in Minecraft, weapon skins on an FPS game, or in combination with other unrelated NFTs. Further ideas and example use cases can be found on the main site. 

Artglixxx

ARTGLIXXX is powered by GLICPIXXX. It extends the concept and ecosystem of generative, algorithmic, programmable art by remixing the original designs of the GLICPIXXX collection. The first ARTGLIXXX series is ARTGLIXXX GENESIS, which can be minted by holding an NFT from the GLICPIXXXVER002 grand collection. 

https://opensea.io/collection/artglixxx

$Glix Utility token: 

$GLIX is an ERC-20 utility token for the GLICPIXXX ecosystem. 

Token IDs between 0–14999 of the ARTGLIXXX GENESIS series (aka ELDEST ARTGLIXXX) automatically yield 10 $GLIX tokens per day for holders until 2032 (10 years from the launch date). $GLIX tokens can also be burned (in increasing quantities) to mint or collect upcoming generations (or versions) of the ARTGLIXXX series. 

Further BGAN tokens were created in the ecosystem, known as $BGAN and $Bastard. $BGAN is the token used in the NFTX protocol and SushiSwap, 

With a more mysterious origin, the $BASTARD token was created and airdropped anonymously to all BGAN holders. After thorough due-diligence by (Qexd and Rosco), the community voted to add liquidity to it on SushiSwap. It can also be used for internal projects like minting NFTs from Berk’s copymint collection.

The official $BASTARD address is: 0x78d9a9355a7823887868492c47368956ea473618 

BGAN DAO:

As aligned with any trustless cohort of degens, the BGAN DAO (established 15 Nov. 2021) is governed through a meritocratic system of proposals that align with the constitution. 

At the time of writing, there are currently nine ‘Signers’ (elected by the community) who help execute valid proposals submitted on snapshot.org (here). 

The current signers include: 

  • Wassoshi – is an artist, postdoc researcher at the University of Copenhagen, and musician.
  • Dirtbaglu – an NFT collector and cryptocurrency expert with a focus on DeFi, liquidity provisioning systems such as nftx.org, and modern contemporary digital art.
  • Crypto Texan – is a Decentralized Finance Community & Marketing at @0xPolygon and an experienced podcast host with a specialization in crypto markets.
  • Qexd – is an NFT and cryptocurrency investor with a focus on credit-based stablecoins, and a host of several innovative BGANs Twitter spaces.

The DAO also includes a respected group of community moderators such as MoonOverlord and ShroomDaddy. Further moderators include JU7URNA, Masal, Rooste,r, and Lyssandros. The Bastard DAO is also currently the largest liquidity provider of the xBGANWETH SushiSwap pool. 

Interview with Berk

Given the unique ecosystem and growing developments of BGANs, it was only fitting to interview the creator, Berk, and dive further into the genesis, the creative process, and his extended thoughts on AI, NFTs, and Web3 in general. 

Mason: Why did you choose CryptoPunks for the training data? Would you still use them as a focal point today?

Berk: When I started with GAN, I had just discovered NFTs (around late 2019), and two significant projects caught my interest: CryptoKitties and CryptoPunks. I wanted to train my dataset, and such large-scale (10000+) collections seemed pretty straightforward to work on. 

I had a NVIDIA GTX 1660 Super as the graphics card, which would be like a snail for training GAN on big resolutions. So I had to be pragmatic and create a dataset with the smallest source images possible: pixel art. So choosing CryptoPunks was an arbitrary decision, but when I started getting the first results, ideas for the BASTARDS concept came, and I started minting them to create new personas on interwebz.

Mason: What made you settle on ‘11,305’ V2s instead of the conventional 10,000 collection size?

Berk: When preparing for V2, I wasn’t sure how big the collection could be. I had a model for infinite numbers, where people can concatenate a background-less bastard and a glicpixxx and mint it. But then I got feedback from Discord users that a fixed number like 10k would be better. 

Then 11,305 came kind of arbitrarily. I was curating the collection by generating outputs, putting them in the collection, and vice versa. In the end, when I said I would stop, there were 11,305 pieces.

Mason: What makes BGANs more than another CryptoPunk derivative like V3 Punks or Phunks?

Berk: Firstly, it introduces a new paradigm to PFPs and a narrative/concept aligned with the creative methodology. And in a market where a digital “Punk” became an elite flexing object, ugly productions from GAN (aka bastards without parents) have a new saying, introducing the new generation of people decentralizing this corrupted world. I think the most undervalued part of BGANs is that they are more real and humane than any of the PFPs, which claim to have a worldly character.

Mason: How did you get the ‘HYPED AF’ GIFs to loop so well?  

Berk: It’s not that complicated. Anyone who uses libraries like STYLEGAN can create those by outputting latent space interpolation. However, on V2 specifically, the gifs weren’t looping perfectly, so I had to open the source code and tweak some parameters to get almost perfect loops.

Mason: The term ‘NFT’ has been corrupted by more than a few influencers, celebrities and brands. With that in mind, other than CryptoArt, how would you define BGANs?  

Berk: Bastards constantly want to make the world beautiful. 

Mason: The visuals of the V1 and V2 collections are quite different. Stylistically, you could say that V1 resembles Francis Bacon’s paintings, whereas V2 leans more towards the conventional pixelated design of NFTs. Why is this?  

Berk: I made a mistake with V1s. Normally, CryptoPunks assets are 24×24, and for GAN training, you have to set a database consisting of images of the resolution, multiples of 32. Like 32×32, 64×64, 256×256, 1024×1024, etc. So, I upscaled the images from 24x to 32x, and it caused some distortions on some pixels. So I learned from those mistakes for V2, and they looked more like original CP and pixel art avatars (I think).

Mason: What are some of your biggest creative influences? Is there anyone in AI or blockchain that you recommend learning more about?

Berk: John Cage, John Oswald of Plunderphonics, Nick Briz, Cory Arcangel (maker of Super Mario Clouds), and Chris Novello who created the ‘Illucia’ project using code-bending.

For any artist new to the Crypto and NFT space, I highly recommend  Rhea Myers – a True OG CryptoArtist. 

And the Trash Art movement has been pretty influential.

Mason: How did Mark Cuban end up with several V2 Bastard Gan Punks?  

Berk: As I know, Haralabos Voulgaris, the former Director of Quantitative Research and Development for the Mavericks, minted a shitton of BGANs (and dumped most of them afterwards, lol) and told Mark Cuban about it.

Mason: What inspired you to start programming and move into AI, and do you have any advice for readers who want to learn more about it?

Berk: I have a piano and classical composition background, so when I saw the possibilities of programming in electronic music, I dived into the subject. Then I took a Python course and followed some other courses online. I realized that I have a super-powerful tool with programming that can do enormous tasks and experiments I can’t do myself. When I was doing my Master’s degree at ArtScience Interfaculty, I was genuinely interested in machine learning methods, so I started playing with object detection, natural language processing, GAN, audio datasets etc. I didn’t follow a specific source or book. I just jumped into everything I found interesting and was capable of fiddling with.

Mason: Are you working on any other projects?

Berk: I am currently trying to utilize Glicpixxx for fashion textile experiments. So let’s see what kind of stuff this will create!

And there is a multiplayer art experience I am working on with a small group of people, which will utilize Glicpixxx and possibly BGANs in a weirdo way. Other than those, I don’t have specific projects I am working on. Ideas come to me spontaneously, depending on the moment.

Mason: Are you proud of any particular moment in the BGAN timeline?

Berk: When I reversed the bonding curve and sent all the minting price to donations, we managed to contribute to saving a baby with spinal muscular atrophy (alongside other charitable causes).

Mason: Lastly, what would be your ideal future for generative art, NFTs, and artificial intelligence?

Berk: I think landscape needs more fun, goofiness and weirdness. Of course, I am always excited to see new technology utilizations. Still, sometimes NFTs are utilized to satisfy the collectors, to make their digital objects ‘worth holding something’. Now NFTs are a part of this world, and we will see so many new uses of digitally created objects, so I am excited to see what kind of craziness we will witness.

Conclusion:

With a pool on NFTX funded with over 10 ETH, a coordinated DAO and Treasury, more than 2,800 unique holders, and over 8,200 volume in ETH on OpenSea, BGANs have gained a reputation as something of a stablecoin equivalent in the ever transient landscape of NFTs. 

But with an eclectic community of technologists, artists, and musicians that form the collective mosaic of bastards, perhaps no better title could be had. Especially when, regardless of ETH prices, there will always be a band playing for the fam in BGAN land.  

Links: 

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.