Using AI to create a King Arthur picture book

About a year ago, disappointed by the lack of illustrated, child friendly versions of the classic tales of King Arthur, I rewrote The Sword and the Stone for my daughter. I had planned for it to be the first in a six part series that told the story of Arthurian legend stretching from Arthur pulling Excalibur from the stone, to him recovering the sword from the Lady of the Lake, his love story with Guinevere, Lancelot's role, and of course the quest for the Hold Grail. Finishing with his death, and the long wait for his return.

At the time, I thought about hiring an illustrator so that I could produce an .epub version that my daughter could read on her iPad, but by then she had swiftly moved on to (checks notes) something called Magic Mixies, which seem devoid of imagination or intrigue and may rival the Tron remake when it comes to soulless energy, but I digress. Around this time I left my job to join an early stage startup and the Arthurian project just fizzled out.

Fast forward to present day, and its really hard to ignore that AI is currently at the core of all positive and negative stories in the tech community. ChatGPT and Google Bard are early frontrunners on the messaging side while Dall•e and Midjourney seemingly lead to way on AI generated images. We're in those heady days where the possibilities that the technology can unlock seem limitless, while the number of jobs that it will potentially replace also seem... well limitless.

With this as the backdrop, I began to ponder whether AI can solve my problem regarding illustrations for my King Arthur book. Firstly I'd like to know, can a tool like Midjourney create reusable characters, and it turns out, it can. This is indeed a promising start, but can it create something that both satisfies my love of Arthurian legend, and is also appealing to my six year old daughter? The answer is undoubtedly yes. I start by asking Midjourney for an Arthur in his 50's, long blonde beard in the Pixar style, as you see from the results it performs admirably.

Playing around a little, I see if it can represent Arthur using a Aardman stop-motion style, famous for the Wallace and Gromit films. Again, I think it does a pretty good job.

Disclaimer: At this point I want to mention I side strongly with content creators regarding how these AI models are learning from, and using people's styles without paying royalties or copyright infringements. What I'm about to do was mainly because I wanted to see if it could be done, and not because I plan to make money or benefit in any way.

My daughter's favourite author and illustrator are Julia Donaldson and Axel Scheffler. If you have children you might know them for their prodigious body of work including the Gruffalo, Zog, Room on the broom, The Smeds and the Smoos, and many more. I really wanted to know if I could create a book that rivalled that level of illustrative quality.

Generating a picture of a sad looking King and Queen outside of a castle wasn't as easy as I'd hoped and took a surprising amount of trial and error around the query language to feed into Midjourney. Eventually this command landscape image of a castle in a green forested area, outside are a well dressed king and queen, in their 40s, holding each other and looking sad and upset. Cute image in the style of children’s illustrator Axel Scheffler produced the following set of images.

This was great, and also really scary as, at a glance, you would think that they were Axel Scheffler illustrations. That said, one issue I had was the shape, I wanted a truly immersive book, and to do that the image couldn't just be square, it has to span the entire width of the two pages. To achieve this I jumped over to Dall•e and used it's excellent generation tools to double the width of the image.

Look at how, with very little effort, I was able to match the style and colours of the illustration and extend them to a second page.

No Arthurian tale is complete without the powerful magician Merlin (Myrddin in Welsh) so again I turned to Midjourney to help me out. After a couple of failed attempts, I finally settled on a version that I liked. In discord, adding the reaction email produces the seed ID, you can then use that value to reference this image in future prompts to create the consistent characters required for continuity in books.

If you know the source material, you'll notice that while each of these illustrations looks like an Axel Scheffler drawing, they don't all have the same style, as though they've been taken from separate books. For my purposes, that's totally fine. It eases my conscience a little and its unlikely to be something that my daughter will pick up on. If I was really trying, I could spend a lot more time perfecting and getting the styles all in sync.

Final thoughts

This was really just an experiment that I thought might be interesting to share. AI's potential is frightening, both as a force for good and for bad. We'll have to see which path it takes, but in the meantime, I'll have fun experimenting with using it for little things like this that put a smile on my children's faces.

Here's the first three pages of my version of The Sword in the Stone.