Artista ex Machina?

Oh, what’s this? Microsoft Bing AI now includes a free “artificial-intelligence” image creator powered by DALL-E? “Sorry, mate, couldn’t resist”, to quote Jack Sparrow from “The Curse of the Black Pearl”. Hehe. Anyway, for the first time I’ve really sunk my teeth into playing around with “AI art”, and the results are…interesting. I first caught wind of it on (where else?) Twitter, where I saw a user generate a rather great-looking set of log-cabin-style apartment developments.

So, inspired by that as well as my own idle thoughts concerning a futuristic far-flung SoCal exurb, my first foray into AI art was the prompt “Log-cabin-style suburban neighborhood with broad pedestrial-and-bicycle-only cobblestone street set in the Eastern Sierras of California”. Well, get a load of what it gave me: worthy of McMansion Hell, this is! Sure, it might very superficially look good, but a second’s glance will reveal materials warping, dimensions somewhat wrong, roofs blending into each other…almost like something out of a nightmare. Just a bit disquieting.

Despite specifying that I wanted a pedestrianized street, it gives me a vehicle that looks like a creepified version of a Mars rover, one of those Space Exploration Vehicles NASA’s prototyping. And that doesn’t even get into how the bicycles are a jumble of warped wheels like something out of a surrealistic junkyard. The US flag is also warped, but in all fairness that is how suburbs often look. Oh, and by the way, the houses don’t look much at all like log cabins.

It does nail the creepy vibe of the Owens Valley, though (source: I’ve been there), and the backdrop is a good rendition of the Eastern Sierras. The cobblestones, grass, and sidewalks don’t look bad either. I generated quite a few other images along similar lines, but this first one was the best.

My next foray was into a more happening part of SoCal. My prompt? “Looking out from the balcony of a swanky upscale house in the style of Frank Lloyd Wright toward the Pacific Ocean at sunset at Malibu Beach with futuristic Coruscant-style flying cars passing by overhead, in the style of William-Adolphe Bouguereau”. Sounds awesome, doesn’t it? The reality? Meh at best. It looks nothing like anything Bouguereau would have rendered, and the flying cars look awful, as does the beachside neighborhood under the house. The landscape is passable, but it doesn’t look anything like Malibu. The house in the foreground does give off a Star Wars concept art vibe, though. All told a real artist could make this scene into something cool and beautiful, but it’d take a lot of work.

I then played around with a nighttime futuristic Pacific landscape, and this turned out a lot better, though still “eh”, aside from maybe the light streaks and most of the sky. My prompt added that it was a “star-studded night”, and the AI rendered that gorgeously. I also wanted “futuristic Coruscant-style lanes of flying cars passing by overhead”, and it just gave me isolated light streaks, nothing like the 3D lanes of solid vehicles passing by overhead.

It also added on a lot of ground traffic, which I didn’t want; I tried specifically telling it to not do that, but to no avail. Indeed, this picture might well be the best I made in my go-round yesterday afternoon. Oh, and needless to say it looks more like something from a latter-day sci-fi movie than anything remotely resembling Bouguereau’s or Merle’s paintings. It doesn’t even look like a realistic oil painting! I’d do better with the oilize filter on paint dot net!

Another image it rendered was somewhat better with the streaks, and the sky and water were pretty, but the rest (especially the overhang) was…eh. I then put it on the shelf for a while.

Then last night I was inspired to come back again: if it looks like a modern-day sci-fi movie, why not go all-in on that? Why not go all in on the Simcity-esque houses that populate the SoCal coast? So I prompted it “looking out from the open balcony of a swanky contemporary house in the style of ‘Tron: Legacy’ looking out on the Pacific Ocean with a star-studded night sky”, and I must say, I actually like this result! Sure, it has its jankiness, but it wouldn’t need much of a touch-up to be an awesome artwork. The vibe and mood especially of what it can generate with a dark night sky and the water is actually pretty.

Below is another generation with this prompt that I liked.

I then added to the prompt that I wanted such a house to be looking out toward the coast mountain range next to the Pacific Ocean, and the result, as I expected, does look quite Californian. I really like the angle it generated, and the composition isn’t bad. It looks a bit cold and soulless the way it rendered it, but a bit of touchup work coupled with perhaps putting a pretty girl in a futuristic Tron-esque ballgown on the balcony clinging to that pillar, gazing out at the water and the stars, and it would look great.

This one looks a bit…stranger (that shape at the end of the overhang…), but I still kinda like it. Could be a solid base for a more human rendering.

I then got the idea to take the house out of the picture, and just have the shore, the sky, and a pretty girl…kinda like some of my own artworks. I got to wondering how well DALL-E could imitate my own style, since it probably scraped all of my artworks as part of its training. Well, it might have tried, but much like Zachary Smith in “Lost in Space” it seems it’s impossible to make a duplicate of me. Hehe.

I mean, sure, the generation below has its pleasant qualities, but does that look like my artwork or my style? Uh, no. I would say my job is safe for now, but I’ve never made a cent off any of my artworks (nobody likes them I guess), so…eh.

I then removed the requirement to emulate me, and ironically the result’s actually closer to something I’d make. I actually really like this generation of a beautiful girl looking at a dark star-studded night sky over the ocean. I detect a hint of a romantic quality to this scene…

…Which then got me to thinking about a more atmospheric generation. So I specified it had to be a “feminine” girl “gazing over the crashing surf” with a “soft ethereal romantic atmosphere”. Most of the generations were hideous in one way or another, but the one below was dreamy, a lot like what I had in mind for this piece.

DALL-E’s beautiful girls tend to be one-note, sporting long brunette hair and thin dimensions along with formal party dresses, so I tried to change it up a bit by specifying a “voluptuous girl”…and it gave me more thin girls. Though this one has more luscious and shiny hair, a softer body, bigger buttocks, and more of an hourglass figure, so it’s not all bad. The surf looks more ethereal in this version, as does the darling’s dress; I swear the thing’s sheer and I can see the outline of her buttocks, but so much the better for a night of sea and spray where the air is warm and sultry (or as sultry as it gets on the Pacific, anyway; the humidity’s peanuts compared to the Gulf of Mexico…).

There’s also another generation I liked, though not as much. The girl looks…weird in terms of the dress and the way her body is rendered, but I quite like the pose of her resting one arm on a rock high above the crashing surf. I stopped my efforts there last night.

Then this morning when I was getting ready to make this post I thought “well, why not use an AI generated image as the featured image for this post?” The perfect thing was to take the theme I liked best, namely the girl on the surf, and make it clear it was about art. Most obvious way to do that was to depict her painting at an easel. Most of the generations from DALL-E were truly hideous and inhuman in some way, but the one below was closer to what I was looking for. Again, her body is strangely rendered, as are the wood pillars for the easel, but I could see this becoming a great scene if a human artist re-rendered it more manually.

But the best generation by far was the one below. Just dreamy…except for that bizarre easel/paintbrush/knife/??? thing that looked like it was stabbing the poor darling right in the chest as it floated beside her in silhouette. Most disconcerting.

So I went into paint dot net and just manually stretched the neighboring pixels in and digitally airbrushed it so it wouldn’t look like an abrupt line, instead blending in with the ethereal surf. So there you go, the first Adamas Nemesis featured image generated with the help of “artificial intelligence”:

Oh, and because it would be just too appropriate to resist, I made the title of this post, which ended up being “Artista ex Machina”, with the help of “artificial intelligence”, in this case ChatGPT. Below is my conversation with it.

Quite helpful, that bot. Back to Bing, it’s not so helpful when it comes to having it write my blog posts for me. See the image below for the result when I tried that, just out of curiosity. Sydney might be a good Bing, but she’s not a very good Adamas Nemesis.

Once again, it seems duplicating me is beyond the realm of technical feasibility…at least for now. Just wait until you have a whole army of automated mini-mes to contend with!

Anyway, the rather unimpressive, if still interesting, results I’ve shared with you hook me enough to want to look into AI art further. I might want to withhold judgment until then. Nevertheless, even the best results I’ve seen other people get with these tools, the nature of the process of generation, and my own temperament and goals in art all give me a hunch that I won’t be giving up my stylus, my tablet, or my camera anytime soon.

Leave a Reply

Your email address will not be published. Required fields are marked *