Delightfully fictional

Midjourney Art for Authors

Since generative art is currently creating quite a stir, for the past several weeks I’ve been playing with Midjourney to assess whether it can become a useful tool for authors in the future. For my tests, I tried to create some environments and characters from my steampunk trilogy The Cerulean Airship. Here are my findings, along with some tips and tricks that might be of help.

Overview

I first became interested in using AI generated images as an author after attending the Self Publishing Show Live earlier this year in London and hearing Joanna Penn talk about the creator economy and Dall-e (which, btw, is now available for the public without waitlist – I haven’t used it yet). A bit later, I joined Midjourney on Discord and the fun began.

First time users receive an initial credit of 10 images. After that, the basic subscription plan (which I have) is 10 USD/month. Beware though: both the 10 images and the monthly credit can vanish quite rapidly. But more about that later in this post.

midjourney art
Steampunk environment

Can I use AI instead of a cover artist?

In short? I don’t think so. Unless you possess some good Photoshop or Clip Studio Paint editing skills, have some knowledge about working with Amazon and IngramSpark cover templates, and know what you are doing.

Though Midjourney can be an excellent tool to generate a nice image to your liking, as an author you still need to employ the skills of a cover artist. The reason is pretty simple. A cover is not just a standalone image. It involves typography and visual adaptation of that image. However, with a bit of creative vision and typography skill, any author could manage a nice e-book cover based on AI (the paperback format is a tad more complicated for many reasons). I used such an image to create a cover for The Prodigal Twin, a free novella set in the steampunk world of my trilogy. And even for that I asked for a bit of help from my cover designer. However, it was a start and I liked it.

Covers aside, you can generate wonderful images based on your books, so AI generated art can be great to showcase landscapes, cities, places, buildings, even characters from your book. Which means it can be a valuable tool for book marketing.

What is AI perfect for: environments

When I played with Midjourney, I generated two types of images: environments and characters. The environments were far better than the characters, and the remaster feature brought them to the next level. I will drop a few examples here. On the left side are the upscaled environments without remaster, and on the right side are the same environments remastered. I preferred the remastered ones.

No matter how good, AI is still AI and prone to errors. So, to get a usable image, you still need to make some tweaks in an image editing software. I use Clip Studio Paint (which is very similar to Photoshop), and here are some examples. Left side: remastered unedited AI generated image. Right side: the same image after I edited it a bit in CSP (got rid of extra elements in the image). The good part is that you don’t need advanced skills to do that.

Characters

Here, it gets tricky. One of Midjourney’s main downsides is the lack of consistency (which is normal in these rather early stages of AI generated art). It never creates the same image twice, so the possibility to create an original character in multiple postures and environments is null. You can create characters, but you can’t replicate them. However, it can give you some nice options to see how your beloved characters might look like. Choose your favourite and show it to your fans.

Unfortunately, the only way to keep character consistency is using a celebrity name in your prompt, which is something I would never do, for reasons I will explain in a moment.

Whilst for landscapes the remaster feature does wonders, unfortunately I couldn’t say the same about characters. It was really hard to get a character close to what I had in mind, and for some characters I preferred the upscaled, non-remastered version.

Some examples below

Simple upscale
Remaster

Midjourney doesn’t excel at…

Dynamic environments and scenes featuring a character within an environment. Unfortunately, even though Midjourney can get you a breathtaking landscape or a nice character, it very rarely does both. I managed to get a nice image only once (the Victorian woman below). It either gets you an image with a great background (but deformed characters so flawed that you really need to be an artist to make them usable) or gets you an image with a nice character and almost zero background, regardless of your detailed prompt instructions for both. But never fear. As it improves with lightspeed, I’m pretty sure this will change soon.

midjourney art

On to the ethical part

Although it is a great tool that everyone can use, Midjourney (and AI generated art in general) is under scrutiny and currently shrouded in controversy. However, I think a compromise can be reached, and there is such thing as ethical usage so long as certain boundaries are understood. Since it’s a highly controversial topic, I would like to underline that what I’m writing here reflects solely my own opinion, my own usage policy based on my own principles, and I don’t intend to blame or to judge anyone who sees this in a different light.

I mentioned earlier that one way to keep your characters consistent is to use a celebrity name, so that the bot understands and replicates something it already knows. However, I would never do that mainly because I am not a fan of using someone else’s image and name without permission, for something I created, and secondly because this would take a lot from the character’s originality.

Another thing I would not do is use artists’ names in prompts. I totally understand why many users do this, but I think those artists’ work and style are their own, the result of many years of hard work, and using what they created with their own talent would be a bit like cheating. I think you can still get great results even without constantly using names of popular artists.

Technicalities

Below you’ll find a list of tips that might help you navigate through Midjourney. And, at the end of this post, I’ll leave some useful links.

  • Using Midjourney can be expensive, so be aware of that. To preserve CPU and not use all your quota in a blink of an eye, upscale ONLY the result(s) you truly like and want to move forward with. From all kinds of images Midjourney generates, the initial 4-version one is the least CPU-consuming one. What truly eats out CPU is upscaling and remastering. Upscaling images you don’t really like equals wasting precious CPU.
  • For the reason above, read and learn well how Midjourney’s bot works, what styles are available, how to use them, and how to write a prompt. I’ve studied the tool for about three weeks before generating images. Experimenting when you are not very sure how it works only wastes CPU.
  • Checking other people’s prompts makes research easier. You’ll find the most popular images (prompts included) in the Community Feed of your Midjourney account (browser version, not the app), in the left menu (I am not very sure, but it might be available for subscribers only). Some are absolutely breathtaking. Copy in a notepad the prompts of the images you love, find common patterns, and learn from them.
  • Always check the announcement section on the Midjourney Discord server. They are constantly testing new features.
  • Learn the “- -“ parameters, as they are of great help. I’ll leave a link to a great resource explaining these parameters. You’ll find it below, in the resource section of this post
  • Though it’s not a visible option, you actually can delete images from server (after you save them). You only need to click on the Add reaction button on the top right above your image and then choose the red X. Clicking on it deletes the image, and no one will be able to see (or save) it.
  • Almost always more than one roll is needed once you find a basic image you like. Upscale it, make versions, remaster it until you are satisfied with the result.
  • When writing prompts, be careful to feed the bot the right amount of information. Too little and you might get something too undefined and general. Too much and you’ll get a contorted image, with too many details, because the bot doesn’t understand what you want. Keep it clear, simple, concise, and use parameters.
  • Learn how to use MJ’s parameters. They do wonders, I promise. From generating the image at the ratio you want, up to eliminating an element you don’t want in your image.
  • Using the same parameter twice in the same prompt is not a good idea, as it confuses the bot. I used the parameter “- -no” twice in the same prompt and I got something vague, colourless, and unusable (and no, I’m not talking about the image below).
midjourney art
Steampunk London

Wrap-up

So, is Midjourney (or any other generative art software) good for authors? Yes, it is. It can be a powerful ally in your book marketing campaign, or to keep your readers’ interest. It’s amazing that you are actually able to create visual bits from your book, according to your own tastes, and share them with your readers. Offer them visual renderings of the worlds they love, let them immerse in those. As such, book quotes now can have a very powerful ally. Now, the visual part of a book mustn’t necessarily be limited to the cover. Explore the endless possibilities beyond that.

It’s trickier when it comes to covers and characters, but not entirely impossible.

Dabble in it, learn (it’s quite complex), experiment, see if you enjoy it. Most likely, you will.

Resources

Some great websites if you want to learn more about Midjourney and how to use it:

I'm an indie author, translation industry professional, and former magazine editor, with a passion for steampunk, cyberpunk, and historical fiction. My debut novel Laevium, a steampunk story set in Neo-Victorian London, is available from most major online bookstores.

Leave a Reply

Your email address will not be published.