AI book illustrations (follow-up)

Putting MidJourney to the test

UPDATE 20/02/2023: For an update on my thinking around AI art, go here.

A month ago I wrote a newsletter about AI generated images and how they might be useful for authors - as well as poking at some of the ethical concerns around AI generation. Since then I got into the MidJourney beta and have had a chance to try the technology for myself.

If you missed the first post check it out here:

AI generated art for writers

If you’re a regular reader of this newsletter you might have spotted some of the images already. The results I got with no experience of prompting the system are fascinating. With very little effort, I ended up with images that I could happily use to illustrate Tales from the Triverse each week (I’ll get to some of those ethical concerns again shortly).

Take a look at this, which was from a prompt to show a spaceship crashing into a lake:

I find the balance of the image satisfying. I love the turquoise-illuminated water beneath the crashed spaceship, suggesting there’s more ship below the waterline, or possibly some sort of energy leak. The ship design looks like a smashed-up Chris Foss. There’s something dark and mysterious on the shoreline in the background, perhaps a building or another ship. There’s a glimpse of another ship in the sky. I’m intrigued that the entire image has a sunset lighting. There’s a definite sense of atmosphere.

In other words, I was startled by the amount of apparent storytelling going on in this image.

My next experiment was to generate some spaceships in orbit. This is something that I could use in several chapters of Triverse. Results here were far less impressive. They have a bobbly AI image look and are unsatisfying in both composition and concept:

This in particular is weird but not in a good way:

There’s a university in Triverse that has lots of towering spires. It’s loosely based on the real-world Fountains Abbey found in Yorkshire. Attempts to create a cathedral-like university were pretty successful:

I love the painterly style of this. The sense of a massive city in the background, perhaps even a city on fire. I love the burst of light from one of the towers on the left side. The intense glow of the windows in the centre. Here’s an alternate take on the same prompt:

This one looks like a cathedral merged with a spaceship. It has an art deco thing going on, especially in the central area. There’s a suggestion of surrounding landscape around the bottom of the image, and perhaps people milling about outside. Again, there are visual sparks in here that conjure up stories in my head. I’m surprised that an AI generated image can do that.

That last image above, in fact, I’m currently using as the image on my Substack landing page. It’s suggestive without being prescriptive, and is generally on-brand. It has a crafted appearance. At some point I’ll make further changes to it to make it more of an original, human-created piece.

Making changes

Using AI generated art as-is can work, and has numerous benefits if you’re tight for time. The lack of actual intent I still feel takes away from the image. I might be the person who wrote out the prompt, but it’s MidJourney that generated the end result. I am not the artist with any of these images.

I’ve started to poke at this notion in my limited capacity as an amateur illustrator. One of the images I was especially pleased with was for a ‘city of waterfalls’. The prompt generated a series of strikingly evocative images, such as this:

I mean, look at that thing! I have no idea how MidJourney is creating something like that, and it could well be riffing on pre-existing works by talented concept artists. But if I was able to paint something half as good as that image I’d be thrilled. Great framing, brilliant juxtaposition of nature and urban landscapes, intriguing suggestions of people and tech in the silhouetted foreground. I love everything about it.

Here’s a variant:

(in fact, we should talk about variants. That you can instantly generate alternative versions on a theme is quite something. If you’re using an AI image generator for inspiration, it’s an endless well)

I’ve started having a go at adjusting the image, and giving it some human-created influence. This is unfinished, and I’m 100% absolutely no painter, but it’s interesting to compare this thumbnail to the AI original:

It’s unfinished in its current form and I’m unsure if I’ll continue it, but what I find especially interesting is that I doubt I’d be able to paint something even this rough without that AI-generated reference. The sense of depth on the image has definitely come through.

Here’s another thumbnail, this time going for more of an ink sketch style:

I think I’d be fairly pleased, as an amateur artist, to have come up with either of those versions myself. I’ll probably continue working on them to see where they end up.

The problem I have is that I prefer the AI version. Hmm.

What about paying an illustrator?

The elephant in the room continues to be the impact on illustrators. Talented artists who make a living with their art. I have no doubt that if I paid a professional illustrator I would get images vastly better than what I’ve included in this post - technically more competent, aesthetically more pleasing, artistically more satisfying.

The problem is that I generally can’t afford to pay an illustrator. Certainly not for a weekly newsletter like this one. A good illustrator will cost several hundred pounds per commission, which is not an option for me. Even if I worked out a bulk order it would still be prohibitively expensive.

MidJourney is £10 for 200 images. Erk. (practically it works out to less than 200, due to the way the system actually functions - but still)

In truth, no illustrator is losing out by me using something like MidJourney. I wouldn’t be able to afford them in the first place. The problem is if the people and organisations who can afford to pay professionals start switching over to cheap automation.

I hope this new tech settles down into being a useful tool: a way to generate reference material without having to rely on generic stock footage, or for talented artists to take and recombine into new material. I can certainly see how it’s a somewhat unnerving time to be a professional illustrator. For my part, I’m now weighing up whether to generate a bunch of new images to illustrate the next season of Triverse…

---

Meanwhile! I have some new ebook giveaways for you to peruse.

First there is the boldly-named Magic is Real, which is full of fantasy romps.

Then there’s the July Sci-Fi And Fantasy Book Fair, which if it were in a physical venue is definitely something I would want to attend.

If you’re after some new books to read on the cheap, check out both of those links.

Right, thanks for reading, everyone. I’ll be back on Friday with a new Triverse chapter.