Kim Jung Gi R.I.P. (and his AI-model is no substitute)
Goodnight, Superani.
A few days ago, Kim Jung Gi died of a heart attack at the age of 47. Kim Jung Gi, also known on the web as Superani, was famous for his large scale public illustration sessions, some of which you can watch on his Youtube-channel. In those videos you can see an illustrator working without any sketches or scribbles, generating an image out of his own mind, transcoding an idea in his head right onto a canvas. His skill in these regards was outstanding and absolutely unique.
With Kim Jung Gi, the illustration world looses one of the greats of the contemporary illustration world and who influenced a ton of people with his passion for style and work.
Jim Lee, publisher and chief creative officer of DC Comics, called Kim "one of the absolute greats" in a series of tweets remembering the Korean artist, who occasionally designed covers for DC series and participated in drawing workshops through the company.
"@KimJungGiUS was a truly phenomenal talent whose pen and brush wizardry captivated and inspired millions of fans around the world," Lee tweeted. "While he drew some incredible comics, it was his live drawing & his sketchbooks about his life, travels and dreams which spoke to me most."
Marvel Comics editor-in-chief C.B. Cebulski echoed Lee's praise: "There was no one quite like (Kim)," he said of the artist, who also worked on Marvel comic covers.
A few days after his death, this happened:
There is a lot to say about this.
While I do think that AI models trained on styles by specific artist will become a commercial product in the form of modular components for ai based illustration software in the very near future, I also think that it’s very bad style to train an AI model on the style of artists who died a day ago. This is just not something decent thinking humans do.
A few weeks ago I wrote about a paper presenting a new framework to think about these cases. In Mimetic Models: Ethical Implications of AI that Acts Like You explore cases where the creation of AI models that act like a specific person can reflect back on reputations or influence outcomes in the job market. This specific case seems to be one of the first cases of what i called a “Pirate Mimetic AI-Model”, where someone just mindlessly trained a model on the work of one person and generated a wobbly, unreliable imitation from it.
I have my suspicions about the motivations here, not to mention the AI art trolls, but I will cut the guy some slack and believe that this was done to honor the deceased artist.
Then there are also people who dunk on this misguided attempt by dismissing AI generated art alltogether as “soulless and cheap (…) next to the real art by the real artist”.
Though i agree with the overall sentiment in this specific case, the aesthetic stength of image synthesis is not the imitation of specific artists (yet). While I can generate thousands of James Jeans in a few hours, they have nothing to next compared to the real thing. This is true (for now).
The strength of these stochastic libraries is not that, but generating unknown unknowns. Its especially the strange mutations and the weird stuff that is unique and interesting about this new stochastic visual style. The uncanniness and the surprise is exactly what makes the experience of AI art distinct from all other art forms, maybe with exceptions for live performances and action painting, where stochastic and random elements go into the experience of the piece itself.
I more and more think about these AI art models not as technologies to produce singular pieces of artworks, but as pieces of art themselves. The latent spaces of every AI model is a compression of symbolic representations into a few gigabytes of data, a technological artifact that we have yet no definitive language to talk about. I don’t think these models are “intelligent” in any sense of that word. They are a new form of cultural technology akin to writing, print or libraries, and in the case of compressed art, they summarize a whole human visual history.
I consider these models themselves a piece of art, done by a whole collective of engineers and scientists, data scrapers and the prompters, the explorers of latent space. All of this is one giant piece of art and we are only starting to explore it. I like the new school tech romanticism this perspective attaches to a debate that speaks about supposedly “soulless” and “synthetic” visual imagery, where actually its a new form of experience that is just at the beginning stages of development. Remember that all of this technology is 10 years old, and image synthesis really started to become usable a few weeks ago.
In one year, artists will be able to license AI modules for Photoshop “in the style of Greg Rutkowski”, and maybe even Kim Jung Gi, too, given that in an interview in 2018, he had this to say, speaking pretty approvingly about technological progress, AI and art:
Many people are talking more and more about the development of AI (Artificial Intelligence) such as Alpha-Go and the influence they will have on our future lives. And the advancement in internet and technology will broaden our ways to express ourselves, and eventually it will have direct and indirect influence in the art realm as well. The art world will be shown in many different forms or in the artworks themselves. I myself have experienced VR (Virtual Reality) first hand. It was a very good experience to me as an artist, and I remember that the audience also seem to be having a good time. The films are also awakening our senses even more and I look forward to their advancement. I believe the development of new and diverse ways of expressing and new forms of art paradigm due to advancement in technology will make our lives more diverse and interesting. And after some time, when people are tired of these things, they can always go back to doing things in traditional format.
I believe, however misguided this attempt at honoring a deceased artist may have been, Kim Jung Gi would have embraced the existence of these image synthesizers which function as stochastic libraries and provide new ways of access to art history.
When I take one thing from Kim Jung Gis work and interviews, then that he loved making audiences experience art. If AI-based systems can do exactly this in new ways, as wonky and unprecise the results may be at this point, he may have liked it.
These models do produce new imagery, new interesting forms, provide new ways to experince art and are, thus, aesthetically interesting. They have their place in the always evolving art space and Kim Jung understood this.
So, goodnight, Kim, and thanks for all the drawings.