Some articles may include Amazon affiliate links. All proceeds go to helping us pay for original stories and to support writers of speculative fiction. Read more here.
Last week, I wrote an article about how to make a book cover with Midjourney and Canva, and posted two videos about the process: an hour-long video where I showed Midjourney to two real artists, and a second, short video, where I showed how to make a book cover in ten minutes.
Well, I was planning to work on my novel today, so, of course, I looked around for something to procrastinate with.
And I decided to take another stab at making a Midjourney book cover. My first two weren’t that hot, because I only spent a few minutes on each one, just to try out the technology.
I was curious what I could make if I put a little effort into it.
I mean, I wouldn’t buy either of these two books:
I don’t have any artistic skills or background. I don’t remember ever taking any art classes. I must have, at some point, in elementary school maybe? But I have no memory of it. I certainly can’t draw.
I have spent quite a bit of time doing newspaper and magazine layout, but that was twenty years ago. And I created the website you’re looking at now, but that’s a different set of skills entirely.
For me, Midjourney was a pain to learn how to use. I still don’t know anything but a couple of commands — “/imagine” and “/relax”.
That being said, here’s what I produced in one afternoon of work.
The wizard covers
I liked the idea of having a wizard on a cover, so I used the following prompt:
- /imagine wizard writing a spell magical epic illustration trending detailed –ar 2:3
The “/imagine” at the beginning there is like “Simon Says” — you have to precede every command with it. The “–ar 2:3” at the end is the aspect ratio for a standard-sized book cover.
The words in the middle are the kinds of terms that might appear in the caption of pictures of the type I’m interested in. Midjourney is trained on billions of images the programmers found online, and it learned about art by comparing the images themselves to the captions that went with them.
When Midjourney didn’t give me what I liked, I hit the “rerun” button. When I liked something, I would hit the “upscale” button or the “variation” button. After I upscaled a picture and it still wasn’t quite right, I randomly would press one of the other button options: “Light Upscale Redo,” “Beta Upscale Redo” and “Remaster.”
When I got an image I liked, I downloaded it to my computer, opened a new Canva workspace, uploaded the image as the background, and added text. Canva is a free online app that makes it easy to create online graphics, Zoom backgrounds — and book covers. Though their pre-made templates are pretty awful.
The above images are 1600 pixels wide by 2400 pixels high.
The title font is “Sigher.” The subtitle and author name is in “Amatic SC.”
The dark shadow at the bottom is a black gradient rectangle — just go to the “Elements” tab in Canva and search for “transparent gradient black rectangle.”
The fonts and the black shadow rectangle are all available for free with the free version of Canva.
However, I did spend $30 on a one-month subscription to Midjourney, and switched to “/relax” mode in order to get unlimited images.
A cuter, younger wizard
These covers look like they’d be good for a children’s book. So I thought I’d make some with a kid on the cover.
I used the following prompt:
- /imagine cute boy wizard writing a spell magical epic illustration trending detailed –ar 2:3
In the first and last cover below, I cheated. There wasn’t enough empty space above or below the boy for the title. So I hopped over to Dall-E 2, uploaded the image of the boy, then used the outpainting function to fill in the space around it.
Here’s how it looks:
The image above on the left is the original image from Midjourney. There’s not really much room above the wizard’s hat to put the book title. In the second image, I zoomed out — used Dall-E 2’s outpainting tool to create more stuff around the basic image. Dall-E 2 matched the style of the original image then tried to add some things that it thought would fit — like the second book. I asked for a starry sky, I didn’t ask for a second book, but well, there you go. I got a second book.
Here’s the final cover, which uses “Sigher” for the book title and “Amatic SC” for the subtitle and author name:
In the two covers below, I zoomed out for the second book cover. Again, Dall-E 2 created some new stuff — it made a lamp I didn’t ask for. But it’s a nice-looking lamp. Good job, Dall-E 2.
I used the “Dekko” font for the first book title and “Amatic SC” for the subtitle and author name. I used “Amatic SC” for the font for the second book’s title, and “Antonio Light” for the subtitle and author name.
For the last set of wizard boy covers, I didn’t do anything to the original Midjourney image.
But if you look very closely, the cover below has a solid rectangle at the top, behind the subtitle. That’s because I shifted the Midjourney image down a little bit to create more space for the text. If I wanted this to be seamless, I could take it into Gimp and blur out the line, or into Dall-E 2 and create some new sky. If I was getting paid to this, I would make it seamless.
I used “Amatic SC” for all the text on this cover.
In fact, you might be noticing a trend here. Picking a cover image is just the start of the journey. You have to think about the composition of the cover, pick the right fonts, do background shading. You might even want to touch up the images themselves, to smooth out weirdnesses that the AI added. So the initial image might take, say, half an hour to an hour to create with Midjourney — and the fiddly bits to make it look right as part of a book cover might take another two or three hours.
For example, I adjusted the spacing on all the text on all the covers. I spread out the letters, tightened up the space between the lines. This is something that I do without even thinking about it, because I’ve spent so much time doing layout and design. If you’re new to making book covers, however, you’ll mostly likely use the default settings. Default settings look good for a regular document, but they look really bad on a book cover.
Then there are the colors. If you drag an image down to leave empty space above it, you need to fill that space with color. You’ll need to match the main image as closely as possible. Canva offers some suggestions, based on the colors in the image, but they don’t always get it right.
Similarly, you need to pick the right colors and special effects for your book title. If you have a complex background, it might be tempting to add a shadow or outline to make your title stand out, but that just makes it look cheap and gaudy. There are subtler effects, like lift, that you can use, or you can try different colors. Again, you don’t want to be heavy-handed and obvious. Just add enough to create the effect you’re going for.
Another thing I do when designing a book cover — and, again, I’m not actually an artist or a book cover designer, so I’m not saying that I do this particularly well — is I ask myself, “Does this look like a real book cover?” If you go on Amazon or another book store, you can see lots of examples of book covers. But if you filter down to some of the less-popular subgenres, and scroll down to the bottom of the recommended lists to the books that have zero sales and home-made covers, you’ll see the difference between those covers and the best-selling ones. The best-selling covers look polished. They have pop. They have stylistic integrity. There are probably classes out there on exactly how to create that feeling. Classes which I haven’t taken, so can’t tell you what the principles are. I go by gut feel, based mostly on the fact that I’ve reviewed over a thousand books over the past two years for our Free Friday series.
And here are the last two covers in this set. I used “Amatic SC” for all the text except the book title on the second cover, which uses the “Sigher” font.
A space battle
Then I had to try my hand at making covers for my favorite type of books — military sci-fi with lots of explosions.
I used the following prompt:
- /imagine spaceship lazer beam space epic detailed dramatic backlighting –ar 2:3
Again, I generated lots of variations, upscaled the ones I liked, then did variations on those until I got an image that had the look I was going for.
For these covers I used the “HK Modular” font for the title and “Antonio Light” for the subtitle and author name.
Clearly, Midjourney is very good at weird-looking spaceships.
My first vampire cover was a little lame, so I decided to make new ones.
The prompt was:
- /imagine beautiful blonde vampire hunter in a dystopian city –ar 2:3
The title font is “Barriecito” and the subtitle and author name are “Antonio Light.”
Midjourney’s original images were cropped too close to the women’s faces, with no room to put titles above them, so I used Dall-E 2 to outpaint again.
I don’t use Dall-E 2 for the images themselves because first, I don’t like their graphics style. And, second, they charge me for each image. And I’m cheap.
Okay, these next book titles are kind of cheesy. But they’re place holders!
I used the following prompt:
- /imagine handsome medieval fighter in front of a glowing portal dramatic backlighting epic detailed –ar 2:3
The main title font is “Black Mango” and the subtitle and author name are in “Antonio Light.”
I’ve seen a few books with emblems on the cover rather than full-sized images. So I thought I’d try my hand at those.
For these, I used the following prompt to create the border:
- /imagine ancient book border –ar 2:3
This is the result Midjourney gave me, on the first try:
I cropped the image and erased the middle part in Gimp. But you can also crop in Canva, and erase the middle by putting a solid black rectangle over it.
Then I used the following prompt to create the emblems:
- /imagine glowing magical emblem 3d
I didn’t use the “–ar 2:3” command this time because I wanted the emblems to be square.
Then I actually had to edit the emblems in an image editing program because some of them had borders around them or other artifacts. Like this one:
I use Gimp for photo editing. It’s the free, open-source version of Photoshop. But I could have just as easily done the edits in MS Paint or online image editors. All I did was run the eraser tool around the sides.
The title font here is “Obra Letra” and the subtitle and author name are in “Celandine.”
Then, for my final set of book covers, I decided to repurpose some horizontal images I’d made earlier for use as website illustrations.
The title font is “Obra Letra” and the subtitle and author name are in “Antonio Light.”
The prompt I used:
- /imagine beautiful woman writing in a magical library fantasy epic trending illustration detailed –ar 2:1
The aspect ratio is different, because these images were originally intended to illustrate my writing advice column from last night.
In the first two images, I used outpainting in Dall-E 2 to create more stuff above the girls’ heads. For the other two, I just left it as a solid background because I’m getting tired.
I think the combination of Midjourney for the initial image and Dall-E 2 for the outpainting was a great combo. I can’t wait until Midjourney gets the outpainting functionality, so that I don’t have to switch between these two apps.
But other than that… well, it’s been eye-opening.
The covers that you see above were all created in one day, by someone who is not a cover designer, using an AI image generator and a free online image editing app.
Now, the ideas for the covers were my own. The AI isn’t going to do that part for you. Yet.
And it often took a few tries to get something that worked. The AI doesn’t have any common sense, so occasionally body parts would be missing.
The first iterations were usually really bad at hands and faces, but the problems would clear up after a few re-rolls.
Is this going to replace high-end cover designers?
Probably not. Many covers already use stock images, which means that the cover designer is just selecting from what’s already there. Creating a brand new image in Midjourney would actually take longer. But it will speed up time for illustrated covers, since you don’t have to hire a separate illustrator to create the image. Cover designers and illustrators are often two different people since these tasks require very different sets of skills.
No AI right now can create a well-composed, good-looking book cover. Yet.
There is also a learning curve. Learning how to write the prompts to get the pictures you want does take time and practice. I’ve been spending time with Midjourney because I’ve been doing tutorial videos — probably several hours altogether at this point. That helped bring me to where I am today, where I could generate these covers.
Someone who only needs a cover illustration once a year would probably be better off hiring someone to do it, or getting a pre-made cover. Spending ten hours to create something usable might not be worth the time and effort it takes.
But it’s only ten hours the first time. After you get the hang of it, it should go a lot faster.
I think this could be particularly useful for someone writing and publishing a lot of short stories and novellas, where they don’t expect to make a lot of money, and don’t want to spend hundreds on each one for new covers. With Midjourney and Canva, you can create a template that you use for all the covers, to have a consistent style across the whole series, and drop in different Midjourney images for each one, then change the title. Make sure to save the prompt you used to create the images, so that you can new ones in a similar style for future covers.
Edited by Melody Friedenthal
MetaStellar editor and publisher Maria Korolov is a science fiction novelist, writing stories set in a future virtual world. And, during the day, she is an award-winning freelance technology journalist who covers artificial intelligence, cybersecurity and enterprise virtual reality. See her Amazon author page here and follow her on Twitter, Facebook, or LinkedIn. Email her at [email protected]. She is also the editor and publisher of Hypergrid Business, one of the top global sites covering virtual reality.