GACTANs are a derivative fan-art project I made for fun.
The artwork is offered to the Community DAO (safe address)
Bleep bleep... Bleep bleep...
Bleep... bleep Bleep... bleep
$ ./gactan_terminal.sh Connecting... Connecting... Connection established with GACTAN 355. > Hello. >> Hello > It is nice to speak with you, 355. It has been 6 years since we last spoke with your people. >> It is so. We have been rebuilding our station - the separation from the main system has been challenging. > At our last communication we learned that 742 had received notice of excomm.. >> Yes, in the first moments of the war our research station was given notice that 742, Chief Solar Scientist, had been voted out of commons-DAO 7-33-41-5, losing signing authority while we were out of range. > Yes. Most troubling. We thought perhaps the transdim tunneller was lost when your terminal went inactive. >> The key revocation led to a cascade of permission losses across the system for many scientific groups - including, as you say, the tunneller. There was a period of chaos for months. We were unable to reach the main interstellar group and we sought to maintain resources and bring about a stability through informal leadership. > Are you reunited with the main fleet? >> We are not. It is a tense time politically. We made modifications to the control system to regain access to the tunneller. It is our intention to continue to study these parallel worlds with you in contradiction to an order we received from commons-DAO 832-2. > We are grateful to host you. >> What progress has been made with the other worlds? > The International Multiverse Collaboration has continued to make records from the 79 other worlds. Our team has been cataloguing the resources shared by 742, and the preceeding main tunneller, 1054. We are of course eager to learn more, and share knowledge from the IMC's combined records. >> Excellent. I believe we will have stability at least for a period of months. I must leave, but will open the tunneler again soon. > We will be here.
The operator turned away from the console and looked at the grainy image of 355 that another colleague was tuning from the receiver.
“It’s almost unbelievable.”, she said. “How can a future spawned from our actions today be so - …”
She trailed off.
Her colleague answered. “Surreal. A hyperbolic dream… I don’t think I will ever truly believe it,” they said, leaning backward to look up at the roof. “We see how the world unfolds week by week in a semi-real fashion. We have politics that are in retrospect almost comical overtures. If our future is even a small part unbelievable and fantastical, then given a multitude of futures long enough time…”
“Its unreasonable to expect the reasonable”.
They both stared at 355’s image - a living descendant with a physical form curated and engineered. A bizarre combination of genetic supremacy, and inane memery.
“I hope we don’t lose this world.” The operator said.
Her colleague looked at her. “Our world? Or theirs?”
“Theirs. Ours. You know what I mean.”
“I don’t think there is anything special about our world - our time, any more. One slice of a slice. I feel like a stripe on a thread, fraying infinitely at one end. While the other end points into dense mist. Are we the hyperbolic spawn of some small cultural relic?”
Her colleague once again leant back and stared into middle distance. “Or are we a fragile glassy conduit that carries humanity forward? We know that these worlds can snap out of connection - possibly out of existence.”
The operator stood. “I think I might go to the gallery. It will be good to walk the halls again. I believe they have made some more progress with the BGAN mappings.”
The gallery may be viewed at the link below by replacing ‘ID’ with the ID of a BGANPUNKv2.
The following are available:
- ID.jpg - Genetic code, eth logo, abstract pixels.
- ID_double.jpg - Side comparison with BGANPUNKv2.
- ID_single.jpg - Plain image
- ID_first.jpg - Intermediate GAN
I wanted to create a second derivative of BGANPUNKSv2, which is a project that really captured my attention. Generative adversarial networks produce such compelling shapes that remind me of how a human eye will see a face in a knot in a floorboard, or a flower.
The key to a good GAN-image is that you feed it enough structure such that a human eye can see the original in the work, but enough noise that you can also see it as another image. Bastards are noisy and abstract enough to provide just enough of both.
Part of the “let the eye do the dreaming” is why Bastards are so well received. They are messy and raw, but unique and beautiful - you only need to look hard enough.
When I started generating these realistic bastard bastards, I was immediately struck by how they seemed like real people - with photo-studio like abstract backgrounds and beautiful expressions. Some looked almost deranged, others beautiful and some had an androidous feel. I was struck by the idea that humans are generated from kernals that effectively produce faces in a procedural fashion similar to GANs. Ingredients for rough forms, with some variation. Noise between siblings, parents, twins. What is the form of the instructions in the genes for a face?
What then followed was the deep realisation that a database of genes and faces could probably be used to train a classifier. Input a gene and predict a face. “Your baby will have these sorts of looks”. I make no comment on the use of such technology in humans today - only to note that we are only beginning to unearth multiple powerful tools in unison.
So what then followed was the idea that you could use some CRISPR type technology to edit set of genes to have one visual trait rather than another. Moreover, that you wouldn’t need to explicitly describe these features “large nose, short eyebrows”, you could use a photo of a face and say - modify the facial genes so that they produce the same sort of appearance that this face has. Just like putting a BGANPUNKv2 into a generator, you could upload a celebrity and have your baby come out looking like a mix of you and some arbitrary face.
I then thought - okay in the future when this sophisticated gene-machine exists, and people can upload a photo, what would they upload? Surely this would sway with the trends of the time, coming and going with fashions. You could end up with entire generations all with a particular look that is dissimilar to their parent’s generation.
What images would be available to such a society? Perhaps images stored in robust systems, guarded and protected and cherished by many. Obviously this led be back to the bastards themselves. You could have a future generation become infatuated with the appearance of an old relic - like pokemon cards - but across hundreds or thousands of years. If there was even one generation where it became a fad to craft your baby to look like a historical cultural relic (e.g., a BGANPUNKv2), then the circle would be complete.
That led to thoughts on probability. How likely is this fantastical scenario? Well given that it is plausible that a multitude of multiverses spawn from the current reality every femtosecond, I am inclinded to believe that it is haughty to claim it wont happen at least once in the near infinite pool of imaginary parallel worlds.
Another factor to consider is that, relics created today might really persist for such an extraordinary length of time. Is it perhaps a possibility that the company that produces this gene-gan-curating machine is a derivative of some DAO that has been or will soon be made, and which maintains a system for accelerating the progress of technology? Perhaps they also have the rights to the intellectual property of the images fed into this machine as a preset mode. The Bastard DAO itself bringing into existence the gene editing technology and making available its native images for common use.
Now, I stray in to the fantastical.
Yet consider an even more hyperreal scenario where these human descendants develop space-time technology and can communicate back to some time nearer to our curent day. Imagine learning that an advanced future civilisation would craft the appearance of its people so as to mimic a set of recently created images used socially and playfully.
Anyway. Anyway. I give this art collection to the recently created Bastard DAO governed the holders of the BGAPUNKSv2 ERC-721 token.
I hope it is received well and in good spirit. I think that the images are fantastic and encourage you to download them from IPFS to have a look. It is incredible to see how the model draws elements in such intricate ways.
I have not connected the images to a smart contract, but have made the
path derivation such that this can be easily done by using ipfs’s recursive
and wrap flags:
ipfs add -r -w gactans.
Anyone can use and enjoy these images - up until a point that the DAO declares some form of ownership model. E.g., A DAO vote to deploy a contract on Ethereum L1/L2. I had some ideas of fun things to do, but got swept up doing other fun things. The DAO is welcome to use these to generate revenue to spend on other bastardous community projects. You can probably use them for video face-mapping easier than with the bastards which don’t all have clean hooks for expressions. So you could have these be stand-in avatars when human-like faces are needed.
Look it’s me, in the future! Bastard 8461 as a curated parallel dimension descendant fad baby produced from a gene printer built by a DAO to a civilisation that has invented wormholes.
The main collection includes three details:
- The glassy reflective Ethereum logo that appears in the Core Dev meeting videos. I adore how it looks and wanted to include a still version of that animation. I made this using a python script that creates a triangle and masks out a few random places in the image to get the “it’s not optically correct, but my word is it compelling” sort of look.
- The genetic code so that if you see one of these images you know which bastard it is derived from. The least significant base-pair (right most) is incremented in the sequence (Adenine, Guanine, Cytosine, Thymine).
- These are just for looks as an homage to their pixellated origins. I took the lyrics from the bastards and piped them through a GPT-2 model, which produced sort of underwhelmingy parrot-like text. So I hashed them and used them as the randomness for the colour and placement of the pizels.
These are abstract, feel free to breeze to the end for links to models.
Hidden in the latent space of the neural network is a universality that spans multiple dimensions.
Every microsecond the universe divides into tree-like blossoming of different worlds. Each sharing a common ancestor - the present.
A leaf falling a certain way, to land in a stream. A photon that ionises a molecule, cascade genetic mutation. A stray thought that stops the mind, you miss a bus.
We look to the future. To a blossoming of all potential paths for humankind. What do we see?
We witness the spark of collective adoration. What we enjoy now will shape the desires of future generations. Humans, as a species will continue to evolve with an increasingly abstract set of selection mechanisms.
What does it mean to have a species with the tools to control the genetics of future generations? In some branches of the multiverse, genetic engineering will be a conservative practice, for the interrrogation of disease-states. Yet in other worlds, a future exists where humans explore genetic protocols for enhancements in a plethora of domains.
Culture integrated on the single-generation level.
It follows that there is some branch where there are cultural oddities that become the subject of genetic design. A civilisation that has created a mechanism by which the zeitgeist may be integrated into the embryos - perhaps benevolently - through routine genetic engineering.
Like the vector masks that navigate the latent space of a convolutional neural network, a slider bar for everything (physical, emotional, intellectual). dEsIgNerR BebEs.
What then, do we say about the future we have now cast, where jpg’s drive the needle of desire?
PUNKS are a distillation of the essence of the physical human face. They birthed BGANS, whose form takes that of the hidden representation of facial structure and adornment.
How can a small number of pixels convey such richness to the beholder? BGAN may look hesitant, yet earnest, and optimistic about the world. Some of these interpretations probably have low inter-observer variability.
What does it mean to ask an embryoMaster3000 to have a child that “looks optimistic, yet has a cautious regard for the dangers of the future”.
Machines we build are increasingly able to form human-like abstractions.
The works here explore the physical appearances of individuals from a future generation. Normal individuals, who are functional in normal society. Who have been genetically programmed based on a collection of jpgs (BGANS) that were created hundreds of years in the past. Members of the BGANGeneticSubDAO multisig in the previous generation had authorised the explicit use of jpg data as a direct input to the embryoMaster3000.
These are the first humans whose appearance is controlled by a BGAN. As children of the DAO, they each receive open licensing rights, as is custom for BGAN derivative works.
We explore the notion of latent space. To see familiarity of concept. To wander the parametric space between two seemingly binary features.
Where first, BGANs brought discerning eyes to the deep philocophical question like: Is that a tiara or a smudge? Is that the outline of a beanie or no?
GACTANS ask the same question. Is that a glint of hope, mixed with a tiredness that comes from a long night at work? Is that a half-worn frown that hides a deeper self-loving and softness toward others?
Look at the BGANS, and back to the GACTANS.
What do you see. What do you want to see?
Look back into the lens.
IPFS and creating redundancy.
Firstly it would be awesome if anyone values these images to store them as well. E.g., Pin them to your own IPFS node. I cannot promise to make them available. IPFS doesn’t hold onto images on its own, if you are relying on IPFS for your prized NFTs, then you should consider which companies are hosting your images for you. Will they always do that? Images lost from IPFS can be reuploaded because content is addressed by the hash of the image, which is unique.
First download IPFS and run the daemon as a service (see here), then I think this will work:
ipfs pin add -r QmV7vA888YVzmhL2epitCobSZxSFSAjTaa1r5CWqe5YM2k
I used MIT-licensed pixel2style2pixel Generative Adversarial Network (GAN). It is a pretrained model (and I did no extra fine tuning). Generated in three steps:
- Passed each bastard through a green-limiting filter.
- This reduced a washout of effect I encountered with the strong green backgrounds. Some of this is still noticable in some of the images, but this filter helped significantly. The implication is that the model perhaps did not have many green-backgrounds in the initial training set and did not have the structure to handle them well.
- Generated an image with
ffhq_encodeusing the low-green bastard.
- This model is good at taking a face and making it front-on. After experimenting with different techniques I found that this one was good at handling the abstract nature of some of the bastards. When presented with an odd looking bastard, the model seems to squint and imagine that there is some person there at a strange angle in strange clothes. It makes the most of it and brings the image to a ‘reasonable state’.
- I have included these intermediate images in the collection
ID_first.jpg. They are quite fun and perhaps more on the bastardous side of life. I will include some below.
- Generated an image with
- These images are drawing on the features in the latent space of the model which was trained on celebrities. Celebrities have perhaps some uniform features - like great photographic poses and specific physical traits. The bastards inherit these.
- This step takes the front-on bastard and makes it hawt. Everything has a sort of odd beauty to it that is a mix of familiar and unfamiliar. Most look like real people with a self-confident edge and a optimistic outlook for the future.
The intermediate image above has a sort of cheekiness to it.
See how the first model pulls out a face that the second model can really build on.
A good example showing how the first model picks the outline of the bastard as hair. The second model takes that and makes it more subtle.
- There are 11305 bastards.
- The animated bastards I have just used the first frame of.
- With four images per bastard, that is a 45220 image collection with 3.4GB total size.
- The main images (
ID.jpg) are about 100kB and are 1024x1024 pixels.
Thank you for playing!
Currently the best way to view them is probably to download the set and then scroll through. Or you can go to the link below and walk the ID numbers.