As if the internet wasn't already bizarre enough, the deluge of weird AI art created by image generators such as DALL·E 2, MidJourney and Craiyon is making things even stranger. From crossbred cartoon characters to surreal food, apocalyptic selfies muppet fashion shows and – erm – people with tennis balls for heads, DALL·E 2 and others seem to be able to create any weird AI art you can describe in their prompt boxes – with varying degrees of success.
Of course DALL·E 2 and other AI art generators don't think of these outlandish things themselves and don't 'know' what they're helping users to create. They run text prompts through the databases of millions of images and captions that they've learned. This means the results are only as weird as users' own imaginations. There are plenty of existential concerns about where this could all be going and what it means for human artists, but AI won't be taking over the world and turning us into slaves yet. We hope.
OpenAI, the makers of DALL·E 2 have recently announced pricing and that they'll be expanding access to the platform over the next month, but for now there's still a waiting list. Want to learn how to make it work for you? See our guide to how to use DALL·E 2. In the meantime, here's some of the weirdest AI art created by DALL·E 2 and other AI art generators so far.
The weirdest AI art
01. The last selfie ever taken
@thesomethingtrippy (opens in new tab) ♬ The righteous wrath of an honorable man - Colin Stetson (opens in new tab)
One of the latest trends doing the rounds on TikTok is using AI art generators to create "the last selfie ever taken". Started by the Tiktokker @robotoverloards (opens in new tab) and taken up by others, the results appear to be suitably apocalyptic showing androids and skeletons posing for pictures amid desolate scenes of a burning planet. Most worryingly, selfies appear to still be a thing in the end days. There's also an even deeper conspiracy theory brewing – just what is the blue cube that appears in several images? It seems that the AI art generator used to create many of the images is Midjourney, which seems to have a dark style in general – see how the best AI art generators compare.
02. Ronald McDonald in Star Trek
A still of Ronald McDonald in an episode of Star Trek: The Next Generation (1987) #dalle2 pic.twitter.com/UhTzAwCPHlJuly 31, 2022
On a lighter note, a lot of the weirdest AI art created on DALL·E 2 has involved putting famous people and characters in unlikely places. And it seems the only limit is your imagination.
03. 'Average' people in 'average' towns
Average French people in average French towns. The architecture looks pretty much what i think it should be. Many flowers as well. Stylish clothes. I can see the guy on #1 smoking Gauloises cigarettes. #dalle2 pic.twitter.com/cDh5bmZlBmAugust 2, 2022
So let's just recall what people look like today, before the apocalypse and before we sent fastfood restaurants' mascots into deep space. Just average people in average towns. Hay Kranen (opens in new tab) asked DALL·E 2 for just that – average people in towns in various countries. But hmm,... we're not sure we'd want to run into these people in our town.
04. Tennis heads
The digital creator Neural Waves (opens in new tab) has found a solution to the difficulties that DALL-2 and other AI image generators have with faces. Just cover them up. In some of his work, that means futuristic-looking Daft Punk-style helmets. In this series above, tennis balls, because... well, why not. Neural Waves is a good example of a creative making AI work for them, describing work as being "co-created with DALL-E 2", not created by DALL-2. Some of the pieces involve the results of multiple prompts that are then combined and edited in Photoshop and Lightroom.
05. SpongeBob SquarePants Godzilla
#dalle2 This is it. This is the best prompt I'll ever be able to come up with. pic.twitter.com/GoxvSIrwW4July 23, 2022
Perhaps Ai should stick to fantasy then. Of all the DALL·E 2 cartoon mashups around (and there are a lot), this has to be our favourite. It's improved by the giving specific instructions for the artist, not just the combination of two characters.
06. A peanut butter sandwich Rubik's cube
Data scientist Max Woolf (opens in new tab) has been testing how reliably DALL·E 2 can follow instructions when asked to produce food photography and to see if it might be able to produce new ideas for food content on the internet. For example to make a peanut butter sandwich into the shape of a Rubik's cube? Yes, it could.
07. Animals live their dreams...
A raccoon playing tennis at Wimbledon in the 1990s #dalle2 #dalle pic.twitter.com/45tBFKJ9ZJMay 14, 2022
There's a whole category of AI images dedicated to anthropomorphism, showing either playing sports, presenting the news, and much more.
08. ... and battle He-Who-Must-Not-Be-Named
Expect every brand marketing team and its dog to be trying to plot a campaign around AI art generators in the next few months. That could get tired very quickly, but some sharp brands have been very quick of the mark. Heinz was quick out of the starting gates, and so was dogsplanet.com (opens in new tab).
The website for dog lovers has used the Discord-based MidJourney to create all kinds of canine creations with some hilarious results – the epic combat between a cocker spaniel and Voldemort being the highlight. If Harry Potter were a dog, he'd be a cocker spaniel, they reckon, and based on this we'd have to agree.
09. The Apple car
AI art generators have opened up new possibilities for fan renders of products that are rumoured to be in production but whose makers don't want to show us anything yet. We've been wanting to get a peak at the rumoured Apple Car ever since we got wind of the project, and the designer and YouTuber John Mauriello (opens in new tab) figured that DALL·E 2 might be able to imagine what it will look like. Of course typing 'Apple Car' resulted in a car that looked like a fruit, so instead, he went for “Minimalist Sportscar inspired by a MacBook and a Magic Mouse, built out of aluminum and glass”.
10. Duolingo trail cam
https://t.co/HhubYUIr4YJune 10, 2022
We finally get to see what everyone's favourite polyglot owl gets up to while were asleep thanks to the Duolingo trail cam. This was actually created in Craiyon, formerly named DALL-E mini, until the makers of DALL-E had words.
11. Greener cities
It's not all apocalypse and anthropomorphism in AI art. Some people are actually putting it to productive use. Yes, really, and good for them. The artist Zach Katz is using DALL·E 2 to show how city streets could look without cars and with added greenery and water features. It's a whole lot more sensible and suggests AI could have a use in town planning. His Twitter account @betterstreetsai (opens in new tab) has gained thousands of followers in just weeks, and he's got a backlog of requests for locations to try.
pic.twitter.com/rEBHoWR7lHJune 12, 2022
Enough of such sensible, worthwhile pursuits. Why use AI to create something useful when you could create a cross between a Pokémon and a dog?
13. R2D2 gets baptised
pic.twitter.com/5V237Z4P1oJune 12, 2022
There seem to be a lot of Star Wars fans using DALL·E 2 and the like, and some of the weirdest AI art pieces have involved characters being placed in unlikely situations, from Jar Jar Binks on Love Island to a Darth Vader MRI scan, and yes R2DR after finding God. This is another one created in Craiyon.
14. Nike Air Sushi being stalked by a bear
got approved for Dall-e 2 today! #dalle #dalle2"a pair of Nike AF1s made of salmon sashimi, in the distance a hungry-looking bear is approaching, award winning product photo" pic.twitter.com/L1QrCjKsyFAugust 1, 2022
We've seen character mashups, artist mashups, what about clothing? In fact, given some of the footwear collabs we've seen in recent years, we wouldn't be surprised if these see the light of day. And just look how much the bear wants them.
15. The Muppets gatecrash London Fashion Week
⌨ "the most ridiculous looks from london fashion week modelled by the cast of the muppets, lfw"💅#lfw #fashionGenerated by AI with #dalle2 🤖 #dalle 🎨 #aiart pic.twitter.com/Slu25GPTGUJune 6, 2022
And speaking of fashion, we're loving these catwalk look modelled by some of Jim Henson's finest as interpreted by DALL·E 2. The weirdest AI art fashion collection we've seen yet.
16. A man kissing a blue deer
Bwahaha!!! THIS IS SO WEIRD!!! 😳😱🤣💀 SO, for this round in Dall-E 2, I wanted to generate an image of a man kissing a blue deer (which WAS supposed to be anthro, like Bam) in an enchanted forest, painted in a Romantic era style (think late 1700s-mid 1800s). Dying of laughter! pic.twitter.com/d2Ygwzn5OrJuly 31, 2022
If you're using an AI art generator like DALL·E 2 you'll soon learn that you need to be very specific with your prompts. This Twitter user appears to have intended to create an image of a man kidding Bam from Animal Crossing but asked for a man kissing a blue deer. So what they got was, well, a man kissing a blue deer.
17. Borat as Dr Strange
some dall-e 2 results pic.twitter.com/6r28Te1ksLAugust 2, 2022
AI art generators could prove to be an interesting aid for casting directors. Wondering if a certain actor might fit a role? Borat as Dr Strange, for example? Just ask DALL·E 2.
Is AI art really art?
The big question is, but is it art? And it's a question that's generated even more debate after an AI-generated artwork won a fine art competition, apparently without the judges understanding that it was made using AI.
Some argue that AI-generated art is not really art because it is created by a machine and so lacks an intrinsic psychic meaning for the agent. They say that really, AI is not making art but rather replicating art from the database it's been fed but in new combinations and forms. The AI is not drawing on personal or collective experiences the way an artist does.
That said, the AI art generator is acting on human instructions. So while the AI generator itself can't be described as an artist, the human who writes the prompt perhaps could (which might explain why people are getting so protective of the prompts they used to create an image). In this sense the AI art generator would be merely another artistic tool like a brush or a pencil.
As for the commercial aspect of art, the human creators of AI-generated pieces can sell their work, depending on the platform their using. There is some legal debate about whether images created by AI can be copyrighted, but OpenAI has recently announced that DALL·E 2 can now be used to make commercial work – that is, that the human who instructs the machine can sell the resulting work.
We've already seen some people doing so, selling their work as NFTs. And Christies sold an AI-generated artwork, “Edmond de Belamy, from La Famille de Belamy,” by the French art collective Obvious, as far back as 2018.
Can you tell whether a piece of art was made by AI?
First it was deepfakes, now it's AI art generators. It's becoming difficult to tell if something is real or fake, and if something was created by a human or a machine. With images created by some of the free AI art generators, it's fairly obvious because they tend to contain uncanny glitches, blurs and bizarre effects that no human artist would create (unless they were intentionally trying to replicate the look of an AI art generator – now it's getting complicated!).
Some of the more advanced recent releases like DALL·E 2 can be more difficult to pin down. Yes, a lot of images do still have an uncanny valley look about them, but a lot of the platform's output could pass for illustration or photography created by a human. It may be that the only giveaway is when the artist has left on the AI art generator's signature – in the case of DALL·E 2, this is a row of coloured squares at the bottom right of the image.