AI, EI, O. The story of a children's book, written and illustrated by AI.

"AI is nothing if not a loose cannon. If you’ve ever tried to use scissors with your wrong hand, this is the sort of reckless jeopardy you’re dealing with. You’ve had two tequilas, you’ve broadly got a handle on things, but don’t get snippy about the details."

AI, EI, O. The title works, sort of. I can’t really take credit for it either, as it was someone else who thought of that too. Which pretty much sums up my feelings on the whole AI experience.

My process was short and took less than an hour to complete. I used Chat-GPT to turn this title into a children's story and Midjourney to illustrate the visuals. Both utilise years of hard work and learnings from thousands of people, all to be conveniently condensed down to a few mouse clicks whilst I unwrap a KitKat with the other hand. Possibly the more challenging of the two tasks.

"At the beginning it all looks very computerised, that sort of digital art that’s somewhere between CGI and Photoshop and is generally of a futuristic gel lit Ork in spandex"

The story was created using a simple text prompt; a vague summary of a children's book about AI coming to a farm. I didn’t infer that I wanted such a glowing reference of AI, but it managed to implant itself as the story’s hero, even by line three. Trying to reassert my dominance, I asked that ChatGPT re-edit to make it funnier and rhyme. It promptly did both, sort of.

"“I’d like a picture of a cow please” I begin. A grid of photorealistic cows are messaged back, all with marginal variations like a Warhol print"

Once my written story was broken into a series of digestible stanzas I started my relationship with Midjourney for the visuals. Apart from the $10 entrance fee, the most taxing part of the installation was working out how to write prompts into a Discord server. If you’re a soft-bodied nerd like me you might have some experience with chatrooms, but here you’ve got to fight your way through experimenting “newbie” channels to finally score a private message with Midjourney themselves.?

Once you've slid into the DMs, you can start prompting away. “I’d like a picture of a cow please” I begin. A grid of photorealistic cows are messaged back, all with marginal variations like a Warhol print. From here it's a game of wits, whittling down through overly verbose prompts to get an image to my liking. At the beginning it all looks very computerised, that sort of digital art that’s somewhere between CGI and Photoshop and is generally of a futuristic gel lit Ork in spandex. Not right for this farmyard.?

"This was fun to begin with, as I mused over how Caravaggio might draw a chicken shop in Brixton"

After trawling through the community trying to work out how others got anything other than DeviantArt homages, I stumble across a site about Midjourney AI style prompts. It tells me how to input individuals’ names to insert a particular person’s style. This is where it begins to get a little iffy. As long as there is enough of a reference online to an individual’s work – Picasso, Dali, Keith Haring – then AI can smush something together that’s in tune with that aesthetic without them ever knowing.

This was fun to begin with, as I mused over how Caravaggio might draw a chicken shop in Brixton, or Beatrix Potter might muddle her way through etching my new Grindr profile picture. Architects, photographers, printmakers, fashion houses, illustrators. Why not mix anyone in and conflate multiple individuals? Da Vinci draws a mean nude, but loosen it up, man, let’s get a bit of Charlie Brown in there for lols. AI is made for these sorts of weird concoctions. It doesn’t understand what it's drawing, it's just doing it. “You kinda want this sort of stuff, right?”

Which brings me back to the book. After noodling around, copying and pinching prompts from others (“sharing” as I now call it) I could loosely manipulate AI to output what I wanted. I can firmly suggest an aspect ratio, textures, background colours, scale and framing. All of this is with fairly broad brush strokes; AI is nothing if not a loose cannon. If you’ve ever tried to use scissors with your wrong hand, this is the sort of reckless jeopardy you’re dealing with. You’ve had two tequilas, you’ve broadly got a handle on things, but don’t get snippy about the details.?

So AI drew my farmyard story, and to those with an undiscerning eye, it sort of looks right. Everything is a sort of hybrid mutant, marbled as if the assets have been dropped and hastily rearranged before a presentation. A pig might have the feathery down of a duck, or two curly tails, or a hat placed not on their head like you may imagine, but an inch lower, on its knee for some reason.?

Upon the first glimpses of the render, I have to say it looks pretty hot, but as it comes into focus the middle remains oddly untouched like a pie in the microwave. As I’m unable to converse with AI well enough to suggest that it “figures out what the hell that duckling-tractor hybrid is”, I'm left to roll the dice again and hope the next one is better. Taking only 10 seconds to render an image, and $10 a month for the privilege, it's very hard not to keep gambling.?

To sum up, if you want to ballpark a visual without too much specificity, from my initial toe-dipping, AI is very fun. You have to toss aside your pixel-pushing Creative Director tendencies and just appreciate that sometimes a cow and a duck can be the same mushy animal. “AI, EI, O” sort of sounds like a legit story, it sort of looks like Richard Scarry took LSD, and it's maybe sort of coming for our jobs. For now though, I’m happy walking back to that lit firework, still fizzing in the ground, and having a good look. Sort of.?

Words by Mike Davies, Creative Director.