Who are you?
My name is Ethan, but also go by Smitty or DeeplyDreaming.
When did you start making art?
I’ve been making art since I got my hands on photoshop back in 9th grade when my school gave us a free subscription for it. Started off just screwing around with all the different tools and throwing images together, but started to move into some actual pieces throughout high school.
When did you get into AI art?
I did some style-transfers back in 2019, but it felt more like an elaborate filter compared to what some think of as AI art now. I really got into the ai art scene at the beginning of this year, late January to February when my roommate showed me VQGAN + CLIP. A lot of people don’t think much of it at this point but the fact that a computer could just make an image out of thin air at my request was incredible. Moved into DiscoDiffusion probably towards the middle or end of February.
What was one of your first successful prompts?
My first ever prompt with disco diffusion was just “a cowboy’s rest, trending on artstation” here’s the image. Just fully default settings, but still more impressive than quite some things I’ve made today.
What is the difference between a good and great piece of AI art?
I consider myself a novelty junkie and I get bored quick. So what catches my eye is when someone makes something I’ve never seen before, or something I didn’t think AI could pull off. At this point, I think everyone is well versed enough with the program that anyone can make stuff that’s cool. But those who take it above and beyond with some clever prompts or init images or mixed media and photo bashing. Now that’s what I love. When we’re all pulling from the same database, inevitably there will be some overlaps in how our works look, even if prompting for different things, so I think to really make it your own, you gotta participate in it as much as possible.
What is your creative method for creating an AI artwork?
The way I go about my works is vision into creation. I’ll just kinda be going about my day doing whatever when some idea just pops into my head and I can see it pretty clearly. So from there I just try to describe it as best as possible with words that I’d think the machine could understand. I’ve done some more elaborate pieces with init imagers and such, but these days I’ve been pretty busy and haven’t had the time to really brainstorm a project.
Do you have any AI turn offs?
I’ve seen some AI art pieces that have really moved me and then some that just feel… “mass-produced”. There’s a lot of AI art out there that looks very similar to other users pieces. Part of that is based on the strengths/weakness of each program to be fair. But, sometimes I am bummed out that we’re using a technology with one of the few things we can say has truly infinite capabilities in the variety of things it can produce, and I get frequent DMs of people asking me how to reproduce what some other guy is making. Another thing I don’t like is when I can very visibly see the prompt in the image. After getting pretty familiar with this tech, you can see pretty clearly when someone has wrote “unreal engine” or “artstation” etc. Kinda kills the magic of it when the recipe to make it is pretty visible. To me, a lot of this stuff is like seeing a magic trick. Magic tricks don’t have the same wow factor when you can easily see how it was done. Same with AI art. Being that we’re constrained to words, I think the coolest outputs are the ones that are difficult to describe with words. Producing something that goes beyond the language you fed it. Another turn-off is seeing someone make something that looks like what I made. It’s sort of a strange feeling as I might have tinkered around with parameters and prompts all to produce this idea I’m having, and then to see someone replicate it, either intentionally or unintentionally, it feels kind of weird? I sprinkle hints to my prompts around all the time, I like to share artist styles and some other notable parts. But the few times I released an entire prompt in hopes someone would learn from the structure, they instead went right ahead and copy pasted it, which was a weird feeling. Kills the magic.
What are some words you currently enjoy prompting with?
There’s not too many words I’d say I use consistently, I change a lot up between each run. I was on a big wave of using the word “recursion” in everything, or I was into Dan Mumford’s style for a while. But can’t think of any ol’ reliable words I use. But I did write a section in my guide about what I call “reserved words” which are basically more extreme synonyms for regular words, and because of their unique specific use case, you might be able to more accurately capture something. For example, if you’re trying to make a giant monster in a fantasy style. The word “big” is going to reference all things labeled big. We label a ton of different things as big. But now think of how many things we would refer to as “colossal” it’s one of those epic words I feel mostly end up in cinema or novels or other artistic uses. For that reason, I think you may get more mileage out of these words with niche use cases.
Dust & decay is your first collection on objkt. Can you tell us more about this collection?
Sure, for me dust and decay was an exploration of how all things pass with time. Even what feels like the constants of life, civilization, humanity etc. nothing is really safe from time. I storyboarded this project with my roomate and aside from the works, the description and the narrative is half the piece. I was aiming to tell a story with the project in just a few short images, ranging from the onset of things beginning to lose their structure to entire wastelands of dust. And all the beauty that can be found in between these moments in a way that fills you with something between existential dread and peace in knowing that you are minuscule to the happenings in the universe. Overall, it’s a project I enjoyed a lot, and looking forward to the next.
What are your goals in the AI art scene?
I’ve got quite a few ambitions for the AI art scene. I believe there’s a lot of low hanging fruits in the clip-guided diffusion scene. A lot of easy wins of things we can improve on by just comparing what different notebooks do, trying out some new strategies for different things, etc. I would really like to see that CLIP-guided diffusion somehow reaches a point where it is competitive to DALLE or some of the latent models. I don’t aim to make it look anything like their outputs necessarily, but I think there’s ways to improve coherence, the time runs take, and accuracy of prompts while still maintaining clip-guided diffusion’s awesome charm and trippiness. The thing I like about CLIP-guided/disco is that it clearly draws and learns from real images and art, but it falls short of blending in with human-made creations, and I think that’s a good thing. It has created its own niche of art that I would call “the right kind of the uncanny valley”. It feels weird saying this about AI, but the results feel more organic than other architectures. It often gets stuff wrong, but in such an interesting way. When DALLE2 has a flaw for example, its usually a few pixels that look off or some glitched spots. When Clip-guided diffusion makes an error it’s what I call “confidently incorrect” It may put extra elements where they’re not supposed to be like more than two eyes on a face, but it puts those there as if it totally intended to, those eyes will have all the detail of a normal eye and will be blended in the picture pretty convincingly. I think a lot of people really like the latent diffusion stuff, but more and more people seem to be realizing some of the strengths clip-guided has that others methods are yet to pull off.
What projects are you currently working on?
A lot of tinkering with the code and such, doing some freelance work for a video game company. And had another project in mind I’ve been meaning to get started on, calling it Interplanetary Formal, will let that be open to interpretation.
Where do you see AI being used a year from now?
That’s a really tough question, because almost everyday I’ll wake up, check my phone, and see there’s some new unthinkable development in AI. I mean we’ve already got people convinced they’re reaching sentience, I’ve seen text to video, text to 3d models, text to fluid animations for 3d models. It’s seriously crazy and I can’t imagine what’s next. I can envision like 2 weeks out, at the max. Just gotta try to take stuff one step at a time or otherwise I’d drive myself nuts trying to keep up with everything. But I do sort of wonder. No matter how good or fast we get text-to-image AI mastered, it’s still inherently limited by language. If I explain to you a concept, you might end up with a totally different image in your head just because often words don’t do it justice, a picture is worth 1000 words. I imagine further developments may incorporate sketches or image prompts better? And maybe even further in the future with neuralink or something, we can just have a machine read our mind to make a picture lol.
Is anyone you’d like to shoutout?
Yeah, back on the topic of those who really take this tech above and beyond, there’s some artists I’m really inspired by, namely upskydown and Remi Durant who have both created an iconic style of their own and never cease to amaze me with the stuff they make.
Guess I’d wanna end it addressing the “is AI art art?” question. It is. Why? Because I like looking at it. If you don’t like it and you don’t want to consider it art, then that’s your choice. Classifying what is and isn’t art is a waste of time because we’ll never be able to agree on something. The better question to ask is, “are you or the machine the artist?” Here’s a little thought experiment to work through that question. Suppose we’ve been lied to, there is no AI, every prompt is sent to an artist living in a box who is forced to make the requests they are given. Would you call yourself the artist in that case with that knowledge? Better yet, if I commission an artist to make something at my request, can I call myself the artist? And in a more extreme case, suppose our program was actually just a super fine-tuned search engine that find already existing images that perfectly match your prompt. I think it would be more fair to call myself a heavily involved creative director if I’m just running through prompts. For the more elaborate pieces where AI is just one step in the process of many, I think it’s fair to call myself the artist, or at least one of the artists. I opened up the question to twitter and some people made analogies to the role of a film director, or how Walt Disney takes credit for works of people who worked under him, and years ago, same case with artists and their understudies sometimes. I’ve sort of hit another gray area with this question, but I think its still a conversation worth having, especially when some traditional artists do feel somewhat cheated by this technology. For one, many bring up the point that although no copyright law is breached, the works of many have been used without consent nor compensation. Moreover, this technology may threaten the livelihoods of commercial artists or even stock photo companies. I really do sympathize with that. Beyond how it can change an industry, it also may feel unfair that the products made by a skill that took years to master can now be replicated in under a minute on some programs, and by anyone. The incentive to master a traditional art drops a bit when there’s an easy way out, I wonder how future generations will deal with that. On the other hand though, I also notice that there seems to be a greater appreciation for art as a whole now that so many, who previously might not have been able to relate or participate, can now engage with art. Especially because designing good prompts requires that we research artists and movements, I think I speak for a lot of users when I say I’ve learned a lot about art history as well as some really cool current-day artists I enjoy. As the Ai-art realm becomes oversaturated with images generated by the second, the value of these becomes diluted, and in turn the hand-made will become rare and greater appreciated. Beyond that, while there is a case to be made about AI programs exploring the work of artists, I believe artists can leverage the technology to their advantage. I think the greatest, most creative minds will be the ones to take this technology above and beyond and bring so much more to the table than what the machine can do on its own. Something where the whole is greater than the sum of the parts.