AI-generated text, imagery, paintings and other art are evolving at a remarkable pace. Some open source projects are now capable of creating visuals that while imperfect, are getting quite impressive.
This has extended into practical functionality too. As we recently reported, one open-source academic photo editing AI is already capable of retouching old and damaged photos in just seconds, but as well as or better than a human editor could do in hours.
All of this is fascinating to consider, but it also offers worrying prospects for the future of commercial art.
Nonetheless, maybe things aren’t as bad as they seem at a glance. Considering what sometimes happens when an AI is asked to generate themed photos from simple suggestions, it’s not easy to see these algorithms completely displacing humans quite yet. This applies especially if you specifically don’t want to creep people out, as you’ll see below.
A perfect example of AI imperfection is what happens when one of these programs is asked to render photos from a series of written suggestions. Even if you select a range of descriptive words that seems pretty straightforward and simple, the final result doesn’t always quite match any known idea of reality. If anything, it can sometimes look absurdly grotesque, or even terrifying.
To give you an idea of what we mean, here’s what emerged when we asked several different AI image rendering programs to create specific visuals for us.
The ones we chose to use are Enstil, Stable Diffusion and Craiyon (formerly called DALL-E mini). At the time of this posting, none of them requires any sign-in, account creation or anything else, which is why we chose them. You can just visit their rendering pages (linked to above) and type in any descriptive prompt text you like. We thoroughly recommend comparingall three for your own photographic AI hijinks. There are also other programs to choose from, though they might require a sign-in or joining a Discord group.
Now, see for yourself and decide whether you should be more scared of AI taking over your artistic career, or just of these surreal “photos” themselves.
For all of the following, we started with a simple photography-related prompt that didn’t seem like it would be too hard to render. Our descriptive phrase was “photographer taking picture of model”.
To start, here is a genuine stock photo of exactly this from the free stock site Unsplash:
It has a bit of a hipster vibe to it, but overall not bad, and perfectly human-looking.
Now, here is what our first contender for AI magic, Enstil, burped out for us in three different renderings.
So yeah, Something’s not quite right.
Okay, so maybe the next AI can do a bit better. Here were the results from Craiyon using the exact same phrase as with Enstil.
Did you also feel a small part of your brain melt just now? Craiyon makes Enstil look positively sane in comparison.
Finally, here’s Stable Diffusion’s effort at making photos out of that same simple phrase.
Okay, so a bit less bizarrely alien, though that last photographer’s floating lenses have a charmingly lovely poltergeist feel to them.
For a bit more fun, we also tried the phrase “person standing at edge of ocean cliff, foggy dawn, rocks in water” just to spice things up while describing a not-so-uncommon photo composition idea. Enjoy the Lovecraftian dread.
Stable Diffusion (our personal favorite for some semblance of reality)
Overall, I’d love to see what these programs would create if you suggested anything remotely complicated, or just tried to describe literal horror/creepy themes. Even the banal comes out looking like the stuff of deep fever dreams.
To be fair, this AI technology is evolving at an amazingly fast pace, and while it’s easy to laugh at the stuff these programs render, they’re going to improve uncomfortably rapidly and very dramatically. We’d suggest you don’t dismiss AI art or rendered photography too easily if you’re feeling worried about your particular career.
Don't Miss These Articles: