The World’s Largest GIF

Jason Salavon is an artist of contradictions with the undertones of a dark trickster. His work reveals a keen mind for computational systems, but is equally rooted in art history and evolving concepts of image production — a direct line from painting and photography to generative art, and finally, generative AI image generators. His back catalog ranges from sublime abstract visuals to conceptual gambits that prove more poignant upon closer inspection.

I first got to know Salavon’s work in the early 2000s, at a time when finding other digital artists was not a trivial task. I saw that he clearly shared some of the same interests as the proto-generative scene I was embedded in at the time (soon to be centered around Processing), while also recognizing that his practice was referring more to contemporary art discourse than the somewhat insular Ars Electronica/MIT Media Lab futurism that I spoke natively.

Initially, I was attracted to the formalist delights of Shoes, Domestic Production, 1960-1998 (2001), and later American Varietal (US Population, by County, 1790-2000) (2009) — visually stunning pieces that are technically visualizations of the most mundane of data sources. I loved the colors and abstract visuals, but also the implication of an in-joke. Visualization can be accurate, while also being embellished to the point of transforming the pedestrian into rococo.

His “amalgamations” — 76 Blowjobs, The Class of 1988 & The Class of 1967 (1998) and his early viral hit, Every Playboy Centerfold, The Decades (normalized) (2002) — take this data-based approach further, blending it with concerns around photography and memory. These series are conceptually and formally tight — a chef’s kiss blend of pop culture with a note of postmodernism.

But Salavon also has works that are almost nihilistic in their anti-formalism. Spigot (Babbling Self-Portrait) (2010) is an interactive software visualizing Salavon’s Google search history in a blunt form — basic colored squares and the generative faux pas of seemingly random RGB colors (they’re not). 100,000 Abstract Paintings (2001) and Golem (2002) are generative systems, but compared to the current mode of long-form generative art, they render the individual outcome practically meaningless. Here, long-form, becomes a way of emptying the image of value.

TODEM (Tapestry of Decadent Meritocracy) (2023) combines all these threads — data as the framing of a narrative, conceptual formalism and photography as carriers of meaning, but parsed through a mostly automated logic that reminds the viewer that what they’re looking at is a kind of meta-text. The fractalized image grid is at once a compositional device and a parameter of the algorithm, but it also provides a pragmatic marketing device: value per pixel.

TODEM also contains hints of Salavon’s dark humor. The images might be generated from prompts taken from Wikipedia, but they are not neutral renditions of history. There is a late-capitalist Americana being performed here. I can’t resist making connections to J.G. Ballard’s book, The Atrocity Exhibition (1970) and, more darkly, Why I Want to Fuck Ronald Reagan (1968). The generated images seem to reference the last few decades of (gritty) American movie aesthetics blended with 1950s nuclear optimism.

In sum, TODEM is an ambitious project, with generative AI applied as a methodology to create a whole visual world, multilayered fractal rather than a simple image. It is also a perfect extension of Salavon’s work to art on the blockchain, embedding a value matrix and minting logic directly into the work.

Marius Watz: We met once in New York ten or 12 years ago, and I remember feeling a connection to you then, which is why I wanted to write this article. Where did your artistic practice start? Do you come from a family of artists?

Jason Salavon: Yeah, my dad is an artist. He trained as a classical painter from a young age and won a scholarship to the Heron School of Art in Indianapolis. In the 1970s, he made fucked-up paintings that showed his rendering skill, but also seemed really in touch with the counterculture of the time.

My dad didn’t make a living selling paintings, though. He always worked other jobs. That upbringing gave me a feeling for the realities of being an artist. I painted and drew quite a bit as a kid, but I never had a fantasy-driven, romanticized view of an artist’s life and practice.

MW: Where does your own art story begin, then?

JS: In college, I took a drawing class as an elective and completely fell in love with making art all over again, as my own thing. I developed dual paths of studying computer science and studio art. I began using the scripting language HyperTalk on an early Mac, where you could get graphics to the screen quickly. It was thrilling that I could write code to autonomously make things. I would go to bed and wake up in the morning and have a batch of images made by this autonomous system. That was exhilarating. Computation quickly became a completely viable toolset for creating art. I rejected any sense that it existed hierarchically in relation to things like painting or drawing or sculpture.

‍MW: Were you ever interested in generative as a concept on its own? Obviously, you would have started doing this before that word was used a lot. I know I was. The idea of describing generative art as art produced using a mechanism like a piece of software or machine is great, but it says nothing about the intention or the cultural background that you would do it in. Now we’re left with generations of different generative artists who did it for completely different reasons. What was your first realization that generative was a concept for you, and was it important, or is it incidental?

JS: It was fundamental for me. As soon as I got those HyperCard stacks making stuff, I immediately thought of filling space with things made autonomously. I remember having these arguments, and actually still do, about the value proposition: “Are 1,000 pretty good things as good as one great thing?” Is one bespoke thing that a human made by hand better than thousands of algorithmically made things? How do you devise an analysis for that? I still don’t have an answer but, counterintuitively, I was more interested in a machine that could make 1,000 good things than I was in pursuing singular greatness.

What is an artist who works on things that make things? I wanted to be that kind of artist. And I’ve tried to pre-compose my works by asking: “What’s the space I want to build here?” Is it the space of Richard Diebenkorn’s paintings? Is it the space of little superhero figurines? What’s the space you want to fill, and then what are the tools and techniques you might use to fill that space? My spaces tend not to be geared toward abstraction; however, they tend to be geared toward human-relatable spaces of things.

‍MW: You’re clearly systems-driven — at a slight remove from the work. You’re not a painter concerned with making one image; you’re concerned with the systems. How would you describe yourself? Are you a formalist or are you a conceptualist?

JS: This is a good question. I think I have a lot of ’70s-era OG conceptual artist in me, but the work needs to look interesting. So I might be trying to have my cake and eat it, too, with that question. 

MW: You’re a closet formalist.

JS: I think you can make conceptual art generators that create things that also look interesting. This ebbs and flows; some projects lean more conceptual and some might be more formalist. I don’t think these are mutually exclusive and I’m fairly resistant to over-categorization. 

I’m certainly interested in the aesthetics of things, the viscerality of human experience, and the way perception connects to cognitive awareness. What you know about something really influences the way you perceive it. So having a thing that has a naturally embedded narrative about its creation will make it look different. I truly believe this. 

Your perceptual apparatus is constantly changed by the information you have. This is why marketing is such a big deal. People are able to — literally — change the way a handbag looks relative to another handbag just through social mediation.

‍MW: And it relates to the formalism vs. conceptualism question. Your work, Every Playboy Centerfold (1988-1997), the amalgamations, that series seems kind of like a perfect case of this tension in your work between conceptualism and formalism. On the one hand, they’re nihilistic — you’re destroying the image by averaging it. And in 100 Special Moments (2004), which takes people’s literal memories and collapses them, like “you are not special, this photo of you is the same as everyone else’s.”

‍JS: Those were some early, big public successes, and had a lot of traction that was not anticipated at all. I started by taking frames and movie stills and arranging them in grids. It was pretty easy to go from treating a sequence of discrete frames as a dataset to then using a set of genre-specific photographs as data. And instead of making a grid of them, why don’t I just plop them on top of each other and take the mean?

I was excited to make the Every Playboy Centerfold pieces because I didn’t know what it would look like. I printed them myself as photographic mural prints and I remember someone at the lab asking me, “What is that?” and I said, “It’s every Playboy centerfold,” and they just started laughing. They had a moment of like, “Oh, I see that.” It’s just a blob, but one could see the meaning of the blob.

This piece is also a good example of distinctions between abstraction in generative art versus data-driven art which AI dominates now. The “dataset” of Playboy centerfolds is imagery generated from some behavior of our civilization. This seems a far cry from generativity, which invokes automatic processes to create its imagery. For example, Form Study #1 (5,000 units) (2004) was me pseudo-randomly generating profile curves, doing a cylindrical rotation and applying a cool material, all of it automated.

But in my practice, these two approaches have always coexisted. For me, this distinction between generativity and the data-driven is simply a matter of: Are you sampling from the world or are you manufacturing the samples somehow? It’s always felt like dual sides of the same coin, artistically.

‍MW: The ongoing philosophy of the AI scene is “anything public goes.” OpenAI is getting sued for harvesting data without consent and using it in their datasets. As an artist, I don’t think you’re beholden to the same kind of ethics, but even so, AI artists need to have a stated code of ethics. What is your philosophy on this?

JS: That’s a good question. I would never call myself an AI artist or anything like that, by the way. I’ve used public media content since the ’90s, so I’ve been here for a bit. I sample and remix, algorithmically, and I’m obviously comfortable with it.

My position regarding data relates to fair use. You need to have significant transformative actions to differentiate your thing from the source material you’re sampling. If I felt like I was over-leveraging the recognizability or effect of some content, this would feel like an ethical breach.

My version of the ethics of sampling boils down to this: Is the reuse and transformation personal and idiosyncratic enough to separate it and create huge daylight from any prior work?

MW: It becomes the question of: What is the transformation? Is the output satisfying enough? It doesn’t have to be new, it doesn’t have to be unique, but it has to be something that is appropriate and feels solid.

JS: In my mind, the question is: What is the composition? Have they composed something new? 

There was a moment where lens flare by itself was cool enough to put on an image and people would be amazed by it, but that quickly became insufficient. It then became the norm that if you were going to use lens flare, you would want to do it in a meaningful way. I think we’ll look back on the base outputs of diffusion models as similarly outdated and lacking a composer.

MW: One challenge with digital tools and AI is the lightness of creation. It’s interesting to have a system independent of you that can create and still create in your hand, so to speak. 

It’s also challenging, like, how do you get the necessary resistance, not in a sort of mythical “art should be hard” kind of way, but how do you know it’s good enough? What is the tension in that work?

JS: I think that’s a really relevant question to where we are. My answer, again, is that we need to start composing. I agree that people can and should make whatever they want, and the attention brought to digital creativity through the mainstreaming of these tools will be a net gain for everyone.

However, I would be concerned if the talented young artists out there were satisfied with crafting a funny prompt, a clever piece of conditioning data, and getting a nice high-quality 1,024 by 1,024 image out of it. This is not the endpoint. It just can’t be. There’s a real arbitrage opportunity for brilliant young artists to start discovering what the actual art forms are for these tools. 

To me, it will require composition, like stringing together things in time and space, exerting control, and eschewing the instant gratification from social platforms such as “that’s so funny” or “that’s so weird.” It’s more akin to something that a filmmaker does, where you’re gonna have to, you know…

MW: Yeah, you gotta kill your darlings. Leave them on the cutting room floor.

JS: Exactly. We’re at this point where there’s a lot of instant gratification in creating AI art, there’s some degree of self-satisfaction in crafting a better prompt or a more memetic image than somebody else. But great work will not come so easily.

I think the coolest thing that AI art will bring is new forms, hybrid works that will combine, in human terms, something akin to the infinite and the personalizable. But there will be some vision at the helm that’s crafting the experience at a second-order level where the author dictates terms to a model that executes specifics.

MW: Let’s talk about NFTs. What was your first introduction to them and why did you suddenly become interested?

JS: I was aware of them very early, before they were called NFTs. However, I didn't have an idea for a project and wasn't comfortable taking old pieces and putting them on the blockchain. I was already making very large videos with AI models for outdoor projection and some wallpaper pieces, and then it clicked that I wanted to make a very, very large animation in the XY dimension, not necessarily in the time dimension. I could subdivide it into distinct Tiles, and that would be my foray into blockchain. 

It was, as always, an intuitive and conceptually driven framework. Then we got into blockchain and I realized it was cool. Like, tokens are cool, smart contracts are cool, and I could see all kinds of possibilities.

MW: How do NFTs fit with your practice, though? You have a decades-long career of showing in galleries and museums, you’re one of the rare digital artists from the ’90s that managed to bridge the art world gap. I’m guessing you’re not motivated by NFTs as a simple cash grab, so what got you interested?

JS: Art is a place where experimental processes can be pioneered and eventually grow into something mainstream. There’s this possibility to make really interesting work by merging the generative and AI, but these new forms take labor and time. In TODEM's case, I worked on this one project for over a year. 

I’ve rendered 56 versions, each one averaging over a week to produce. Blockchain offers an experiment in creating culture, and a different way to evaluate the success of cultural products, but incentive structures might be counterproductive to making “big important things.” Had I been more plugged into the NFT world, I might have rendered fewer versions…

MW: I really appreciate that you chose to engage with the NFT space not because of the crypto hype of 2021, but because the blockchain offers you protocols and mechanisms that would enable a project like TODEM. From the minting and indexing of media artifacts to the sale and pricing of the same in a global networked market — on custom smart contracts — blockchain is a natural fit at every step. It very much feels like a blockchain-native project.

At the same time, you’ve been open about being somewhat mystified by the crypto culture as seen on Twitter or surrounding people like Beeple, where performative rituals like GM and “to the moon” mantras can feel like secret handshake protocols that are more than a little cult-ish at times.

JS: It’s true. I was unaware of some of the basic conventions. My intentions were around making a thing that was digitally native, and blockchain is obviously perfect for the tokenization of things that are digitally native. But I was unaware of basic things, like how floor price would interact with a non-uniform project like TODEM...

I was also fairly distant from the generative art scene online even though I'd been making the equivalent of generative collections since the early ’90s. “Little Infinity” is a term I’d coined a long time ago for algorithms that create immense variety, but also kind of all look the same — like Fidenzas or Ringers. They’re not big infinities, they’re little. The variation inside of them can’t be that large because the space of possibilities just isn’t that big. And so a Grail inside of that, because you know, you saw the poodle in the cloud, because of pareidolia, that’s a pretty thin way of assigning value. Long term, it’s a problematic proposition.

MW: You’ve described TODEM as a hyper-GIF tapestry, exploring polarity and democratic meritocracy via generative AI and blockchain. 

JS: I call it a hyper-GIF tapestry primarily because it’s 100,000 pixels by 58,000 pixels of ten-second, looping, synthetic animation. I was interested in using AI automation to create something far beyond conventional scale, towards a colossal, near magical scale. So that’s the hyper-GIF element. We began joking that it’s the world’s largest GIF.

MW: And tapestry? Is that a reference to the historical tapestries that kings used to have celebrating their victories, etc.?

‍JS: Yeah. The tapestries in the Cloisters in NYC are an inspiration, like The Hunt of the Unicorn. I’m fascinated by artworks that intend to inspire awe and transport their audience elsewhere. A Vermeer viewed in person has a magical quality even now, just imagine encountering something like that in the 17th century.

MW: I don’t remember the exact quote, but someone said cathedrals are the original multimedia experience. Even visual musical experiences, like an organ, could play colors through candles and control shades of glass. The tapestries we know from history often cover things like war or religious experiences, very much like the sublime. So they did definitely function as a kind of narrative format. 

JS: Yeah, to visit a High Gothic cathedral, even now, produces an awe-inspiring special effects experience. Popular culture has mostly replaced this, whether it’s experiences with film, transportation, television, games, or theme parks. 

I wanted to make more than a big GIF. I wanted to tell a story about Western civilization and harness tools that allow us to access a sort of extra-human compositional power.

MW: That’s interesting. We were talking about this idea of meta art, and this seems like a perfect meta concept. You’re controlling the input, the generators, but you’re also using prompt engineering to influence them in certain ways. 

My observation of looking at TODEM is there’s some darkness in there, and there’s some perversity, which is native to the topic, like J.G. Ballard, the atrocity exhibition, etc., which I find super interesting. To what extent was it planned and orchestrated, or was it random?

JS: A little of both. I’ve worked this way previously, where I pay attention to the samples and get a sense for what the overall distribution feels like, and shape that by imposing my own compositional sensitivities on that distribution, rather than worrying about what any one instance looks like.

I haven’t read all 5,226 prompts, but I did censor the material of two prompts. So I’m concerned there might be a prompt floating in there that’s really offensive that I haven’t seen, and that someone might be bothered by it.

I think this new form asks many questions. At this point, in my own practice, I mostly want to work large-scale, equivalent to epic novels or something. I want to do things that are bigger and take longer. And I'm interested in them being digitally native. 

It would be hard to present TODEM in an exhibition space in this form. It’s too high resolution. It’s digitally native. That’s where it wants to live.

MW: And is infinite in this certain way.

JS: And my next works will get even closer to that. For all the surprises of making a digitally native project and then putting it into an unfamiliar world, I’m already working on the next one. It will be framed differently, but I still think blockchain is the way to make shareable, digitally native artwork. 

MW: What are the next steps for the project?

JS: As ever, I’m going to keep making art. I’m lucky to have some really great people working with me. Everybody's excited about this project and we’re working on new stuff. 

This process taught me a lot and Ivm excited to do another massive one. Even though TODEM is still young, I’m already working on another. The next one will be different, more of a big, multifaceted universe with weirder, in-house AI. That’ll happen sometime soon.