Vadim Epstein

@eps696 ∙ Published February 2, 2024

What’s your background and how did you get into AI art?

The first part of my career was purely technical: a graduate diploma in theoretical physics, then computer engineer and IT consultant for many years. The second part went in parallel, as I applied the obtained tech skills to create some weird stuff. I did it just for fun, yet a few years ago I found one of those early pieces showcased in a huge exhibition in Vancouver alongside works of Picasso, Duchamp, Warhol, to name a few. First I developed a very trippy website, then decided to try digital visuals and got stuck in there for years.

Financial crisis in 2008 had eliminated my nice office job, so I had left it altogether and pursued freelancing in creative industries (being already a leading VJ in the country). I promptly grew to the director of a media production studio, making 3D mapping and interactive installations - that was when I finally got in love with generative practices. Sculpting quasi-organic processes with “quantum” design and chaos theory in hand was an enthralling exercise, allowing artistic reflections both subtle and sophisticated. By 2017 that area got populated too well, so I looked for another fresh & wild habitat among the hottest ones - VR, crypto, AI/ML - and found the latter to be the most promising.

I’m still utterly happy with that choice; artificial neural networks appear to be the new VJing, as they essentially “remix” existing data into something novel, only doing that on the deepest nuclear level. AI/ML integrates the finesse of mathematical abstractions (known from the former procedural practice) with the roughness of real world data, unveiling unprecedented expressive power for complex articulated statements. It’s a proper tool to explore both aesthetic and semantic glitches at once: Computer Vision as the insight, not an eyesight.

What drives you to create?

Creativity is basically a curiosity. The endless “what if”s and “let’s try” can (and usually do) tell fascinating stories with absolutely unpredictable narrative twists.

To be honest, I don’t find much fun in producing complete and well-polished works. The final stage of any serious project consists mostly of tedious curation and technical finalisation to the desired level of perfection; that’s pretty exhausting. But the first stages are really exciting - the process of learning/tweaking/breaking/discovering things is a pure joy, directly from the childhood ::]

What does your workflow look like?

About half of the work is wandering over Twitter and Github, looking for new libraries and methods, and adapting them to my codebase. I do prefer working with my own code and very rarely use foreign ready-made tools, so such regular workouts are crucial for the process, they keep the whole ship afloat. There was a nice blog post by Aaron Hertzmann about that.

The art pieces also usually start from a new tech - a github repo, a fresh trick, or simply an idea. Nine out of ten go to archive or trash, till one may appear promising enough for a thorough investigation. Most often developed techniques end up as simple demos - when I quickly render something just to illustrate the finding while it’s hot, with optional theme and sound for better impression. Sometimes the result happens to be good enough to spend extra efforts for a proper concept, worthy of that expressive discovery. Once in a while it’s so good that it can be employed without any extra concept - that occurred for instance with StarGAN2 feedback loop technique in 2020-2021, bringing over a few months such outcomes as Occurro, Pixie, Ghosts.

The most significant success was development of Aphantasia toolkit in 2021-2022 - an early text-to-image generation technique, a hybrid of Deepdream and CLIP-VQGAN. It appeared to be quite an effective instrument, a year-long affair and a proper novel research, cited in papers (besides an invitation to the Braindrops platform).

In most cases I feel the key is a roaming study without predefined goals, a journey.

That said, there are also complex projects and commissioned works which employ dozens of techniques, and therefore follow classic workflow - from a concept to the product.

How do you imagine AI (art) will be impacting society in the near future?

As long as we keep sticking to the century-old definitions of art (all that stuff about personal will, responsibility and human authorship, not quite relevant these days), AI art (that is, human art employing AI as a tool) is no different from any other artform. And the societal role of any art is mostly education, bringing an opportunity and advantage of getting novel intuitive knowledge about the world.

The famous Stability AI motto “creativity for billions”, bringing unbelievably powerful creative tools to everyone, indeed may have pushed some low-level boundaries for those captured by these exciting prospects. Yet it doesn’t seem to change the whole picture anyhow - what we observe as a result now is mostly countless pseudo-realistic visuals like fantasy scapes with sexy warriors, or desperate efforts to domesticate AI tools by fitting them into Procrustean bed of classic cinematic forms. Very few dare to approach that novel knowledge in its full glory.

As for AI and society in general - there are lots of discussions nowadays with radically opposite views, whether AI would save or kill us. I have very little illusions that adjusting or “aligning” the technologies may bring much good. It’s always a human who is the source of the problems, let’s maybe “align” ourselves first? The progress doesn’t make things better - while one part of the world is embraced with atrocious wars killing thousands people daily, another part spends enormous efforts on discussing ephemeral ethical issues and worshipping their psychological comfort. I doubt AI would help with that.

To speak reasonably about AI impact, we need to step away from the human-centric paradigm first, and consider a wider look on the non-human entities and forces. Until that it looks like a classic wannabe-colonialist approach - to urge and harness things without giving them enough freedom to manifest and emerge on their own.

Who is your favourite artist?

Basically I revere the masters, able to bring forth whole new worlds; absorbing their creations is like getting an extra life. Here are few names which resonated as such:

• text: Vladimir Sorokin, Neal Stephenson, Scott Alexander
• visual: Hieronymus Bosch, Jan Saudek, Anatoly Fomenko
• films: Peter Greenaway, Alejandro Jodorowsky
• music: Psychic TV, Coil, Dvar
• modern art: Jon Raffman, Domnitch & Gelfand
• AI and generative art: Mario Klingemann, Robert Seidel, Roope Rainisto

What is your favourite prompt when creating art?

I’m afraid, there’s no such thing ::] Ok, I’ve used epic cinematic litho prefix for my CLIP-based video tool Illustrip (a part of the above-mentioned Aphantasia), as it always dramatically enhanced style, details and coherence. Few works, employing it: Requiem, Little Science, Naturmort.

That’s probably all, as I prefer exploring technical intrinsics of the methods rather than messing with fanciful wordiness. In general, I see prompt engineering as an interim phenomenon, relevant for the current level of AI progress and probably incited by proprietary closed systems like Midjourney, which cannot be used in full power otherwise - with model fine-tuning, internal semantic interpolations, etc. As such, artificial surgically mastered text inputs are destined to fade out with further advancements into both natural language treatment and mixed media inputs as more flexible and powerful ways of control. If that sounds unlikely for you, try loading fine-tuned Stable Diffusion with a bunch of reference images (e.g. via IP Adapter technique) - you’ll get more impressive and coherent results than with page-long text prompt. That said, I don’t diminish the role of text as a symbolic medium, it’s apparently huge; I do believe though that Math is still a better language.

Do you have a specific project you’re currently working on?

It’s THE POEM, a sequel to my previous biggest project so far, TEOPEMA [2021-2022]; both made in collaboration with musician Ivan Pavlov [COH project].

No surprise, these two are about Language and Math - in a figurative form of course. Main concepts are described on their websites in detail; I’ll add a few important comments here. First part was about creating a visual representation with text and code, exploring the process mostly from the perspective of the final result and using words as obedient tools without their innate significance.

On the contrary, the second part is built around specific text, namely “The Poe-et’s Nightmare” by H.P.Lovecraft (a poetic message from Lovecraft to Edgar Poe). We attempted to “translate” that powerful story into visuals, but not as mere illustrations of the actual storytelling; instead, we aimed to project the original symbolic space of the poem spirit into a new coherent realm, sometimes abstracting away from described events altogether. The core of THE POEM is not only obtained visuals, it’s that act of projection itself - “from the written unseen to the unspeakable evident”.

As our experiments have led to multiple various aesthetics, we have distributed them over a few formats - a film, an NFT collection and a printed book. The production process is mostly completed by now and we’ve started to look for publishing opportunities.

Would you tell us about the AI Surrealism exhibition in NYC?

I loved participating in that affair, cause it had kickstarted our Big Project of the year, described above. The level of that event - diversity, quality of artworks and curation - implied decent attention to all details, so I felt I had to develop something special. One of the two submitted pieces happened to define the aesthetics and technique which complied with the concept which we had in mind for our project.

I also honoured inventiveness of the technical presentation of artworks - such tiled multi-screen setup has been one of my favourite presentation tricks for years too, it has always looked lovely.

What does it mean to be an AI Surrealist for you in the times we live in?

This question pretty much correlates to our choice of the source text for the project described above. Let me quote that part of our concept (with exact terms from the original author):

As Venkatesh Rao points out in his Disturbed Realities essay, largely dedicated to Lovecraft and Ballard, it is imperative to comprehend the collective moods and historic consciousness shifts in the deeply disturbed eras, one of which we’re currently living through. Such a visioner optics can make novel scientific or mathematical knowledge, alongside with other reality expansions, accessible to the less sensitive souls as new emotional range and psychological territories.

The goal or meaning of any art may be to condense such subtle scattered knowledge into comprehensible tangible messages. Since surrealism by definition is about some kind of [alternative] reality, it may have to make it within more accustomed settings, delivering even more obvious and impactful visions.

Would you like to share anything else?

One thing which I always try to pay attention to - some hidden or indirect features of the project. A simple example is easter eggs, adding extra layer(s) of involvement to both creation and consumption processes. Or let’s take a seed value for AI/ML generation; while it doesn’t have any meaning on its own, setting it to some specific value would make you “tune” your mind to that, which essentially makes a difference by changing your own attitude to the process and the product. In short, I prefer acting as if there is some metaphysical judge evaluating all your creative actions - public and personal, obvious and obscure.