Can AI Create Your Next Book Cover?

AI Generated Book Covers

by | Jul 1, 2023

Last Updated:
Jun 25, 2023

You’ve probably read the articles and seen the images because AI image generators are making waves—both good and bad. From an AI image that have won a fair contest, to the controversy of using artist’s artwork without their permission, to an author having to turn down a book cover award when it was revealed the cover artist lied about having used AI generated images, there is a lot going on in the news about AI images.

But man, when you see some of the images AI have generated… well, you see why they do win awards (even if dishonestly winning awards).

And so, you may be wondering if you can use AI generated images in your next book cover—and if you should?! And we’ll cover all of that and more in this post. Along with some great images, of course! 😉 Now let’s put some of these AI generators head to head and see what might help you with your next book cover.

The Controversy with AI Images

I’ve you’ve heard about AI images, then you’ve most likely heard some of the controversy with them. There are a few issues with AI generated images, ranging from an outcry similar to just about every big innovation change to some really key points that have yet to be addressed. Some of the main ones are…

  • Job loss: With every great technology change, there has been a wailing that it will cause job loss. Cameras were going to end painting. Computers would end… well, something. While I’m admittedly minimizing this point, it is valid. There will be artists and designers who lose work and employment over this. If you can generate a book cover for a simple monthly fee to access great images, would you spend hundreds of dollars to hire a designer? As a graphic designer myself, I get this. But I’m also not panicked. These tools are amazingly easy and incredibly frustrating at the same time. I think of them as another tool in my tool bag, just like a new brush. AI might become an element in my designs. I will know how to use it. But it will not replace everything that has to do with art. It still cannot make an actual oil painting or a watercolor. And even with digital work, I’ve not found an AI image that didn’t need some additional tweaks to make it finished and truly stunning.
  • Skill loss: It takes years and a lot of practice to build the skill to create images, not to mention a lot of study of techniques and theory in composition, color, symbolism, etc. When you can create a strong and stunning image with just a few simple prompts, artists will have to wonder what is the point to learn all of those things and spend decades perfecting their technique? Could we have a future where the only “great” artists are computers? I doubt it, but those who master the ability outside of the AI realm will become fewer.
  • Exploitation of artists: In the case of most of the AI image generators, the AI was trained by ingesting hundreds of thousands of images to learn color, composition, what humans like to see, etc. Some of that artwork is from famous dead artists like Monet, Picasso, and Rembrandt. But many, many other images are from modern day artists who might have shared images on social media or their websites where those original images might be for sale or listed in a portfolio. Just like you shouldn’t be taking someone else’s social media pics and using them as your own without asking permission, these images were used without the artist’s permission (or knowledge). Now that this major flaw and artistic “theft” has been called out, most of the creators of these image generators are completely ignoring artist’s pleas to have their images removed from the data banks. While I doubt the model training wasn’t done with the intent to plagiarize other artist’s works, the fact that companies are ignore the issue is condescending and will certainly not make the issue go away.
  • Copyright issues: Who owns an AI image? When a painting is created by an artist, the artist owns the copyright unless they sell it. But an AI can’t own a copyright, so does the person writing the prompt own it? What about the artists’ whose work it is based on? Should they not have a piece of the copyright. Right now, the answer is simply “No.” No one owns the copyright to AI generated artwork and you can’t apply to have it copyrighted. It is in the public domain and so has a creative commons license. If you aren’t familiar with this term, it means that anyone is free to use the image. So, if you generate an awesome AI image, someone else is free to use it as well.
  • Biases: AI image generators are trained on existing images, including a wealth of historical images. Do you know what is wrong with a lot of these images? They are full of white people acting supreme. Other cultures are often minimized and sometimes denigrated. Now, the AI model thinks that is how these people should be portrayed. Pretty sick, right? And I’ve already seen this in action. If you put in a generic prompt like you’ll see in this post, you often get a white girl, usually blonde and blue eyed. Yes, you can ask for a black heroine and will often get a great image, but you have to ask. And sometimes some societal biases will leak through still. Art that challenges or changes perceptions is more difficult to make using an AI image generator. If you want something cutting edge and shifts perspectives… well, you’ll have to make some really fantastically worded prompts, add in elements outside of an image generator, or hire an artist with vision.

There are other controversies out there about AI generated art, but this covers the biggest. You’ll have to decide if you think the impacts of using AI generated images is worth the costs. And be aware of the biases that come out. When you see every suggested image having a white character, take the time to question that. Put in a different prompt to specify people of color or different sexuality.

The fact that you can do that, that you can with the change of a few words ask for a Native American starship captain instead of scouring hundreds of stock image to try to find a photo or hire a custom image to be created, is actually rather awesome and in some ways levels the field. Because, I can promise you, it is hard to find good stock photography of minorities that aren’t already biased (if they exist at all).

Our AI Image Generator Test Scenario

We need a standard to compare these numerous image generation tools. So, let’s say we have a current book series and we are looking to write book 3 and want something similar. So, we have a base look and brand that we want to continue so we don’t have to re-do the previous two books.

You might think this is a harsh scenario. Isn’t it easy to just put in a prompt and make a cover? Yes, but this scenario will give us a few extra things to test for—rather than just generating an awesome image we can slap on a book. Here, we get to see if we can control the branding and feel for a series and author. This is something that we know hiring a book cover artist can accomplish. But can AI stand up to that test?

And it is a real world scenario. With AI generation coming out now, there will be authors in the middle of a series that will have to decide how to round out the rest of the book covers. Redo them all or will switching to AI be seamless?

With that, I’m offering my books up as the experiment and we are going to look for book 3 in this series (and if you are one of my readers, no, I’m not recovering the real book 3 or writing another book in the series. I’m just using this as a test here. 😆)

Scenario for an AI book cover

As you can see, I’m going to probably have a dragon on the cover. I’m doing this NOT just because dragons are awesome (🤣) but because dragons have been notoriously difficult for book cover artists. There are only so many stock photos, paintings, drawings, 3D images of dragons to choose from. Which meant that a lot of book covers used the same dragon images. Will AI allow us to finally have something new? Let’s find out!

Let’s assume I want for this one a young woman hiding from a dragon. I’ll use this as my prompt to test a few of the image generators out there.

AI Image Generator Dall-E 2

Dall-E and the latest version Dall-E 2 are image generators created by OpenAI, which is the same company who created ChatGPT. So, you have to feel a little excited about trying out their image generator. Unlike most/many of the image generators out there, it also allows you to upload a base image.

Does this mean that I can upload one of the series covers, put in a prompt, and presto, have book 3 created with a look that matches. OMG, wouldn’t that be sweet? Well, let’s see.

First off, the layout of Dall-E is very easy to dive into. No confusion here. But we are going to go straight for the “Upload and Image” function, which is in beta phase right now.

Prompt screen for Dall-E

Uploading my book 2 cover with the hopes of using it as a basis for something new, I immediately hit a snag. A square generation frame box appears.

Dall-E frame on uploaded image

My book cover isn’t square and this generation box doesn’t even expand to cover the full width of the cover. Oh, well, we’ve come this far. Let’s see where this goes!

Honestly, not very far. I enter first a prompt to get a sandy haired woman hiding from a dragon. No changes. So I try again with my scenario prompt of a young woman hiding from a dragon. As you can see from the set of images above, I don’t see any changes. Is something supposed to happen here?

Well, a little confused, I decide to just try the general prompt and see where that gets me.

Results from Dall-E prompt first try

Oooh. Just… umm… Yeah, let’s try that again.

Results from Dall-E try 2

So, yay, I did get one minority heroine there. But the rest… I’m not even going to get into the fact that all of these images are square, so completely useless for a book cover because those images… Just no. I think most 7 year olds could handle layering images in Photoshop or Procreate better than this. There isn’t even a dragon in the last two! And the dragons I did get 🤮. And what is with the weird glare in the the two left images just above? The first try amusingly featured books for three of the four tries and the one looks like a tiny segment of an actual book cover… not a very good one though.

Before I move on, I will say that I have played with Dall-E to generate “real world” images. Like someone typing at a computer. Or pencils on paper. Something simple and mundane that doesn’t feature fantasy creatures. In those cases, it works really well. But you still end up with a square image, which is limiting if you want a book cover. Especially with the size that you can download from Dall-E. If you expand the image and then crop just a second for a book cover, you’ll end up with a size too small for most platforms or a blurry book cover.

With Dall-E run through its paces, let’s move on…

Text to Image Generator StarryAI

StarryAi is another AI text generator. I couldn’t find much about the company behind it, how the AI model was trained, or when it was released, but it has a fantasy lean to it and I thought I’d try it out.

StarryAI text to image generator

You start with 25 free credits to use and can get additional free ones for completing prompts such as setting up your account or simply showing up every day. And, of course, you can buy more credits.

To start with, you can choose a style to help guide what your final image will look like. That is a helpful feature. Though, most of the text to image generators will create a style if you type it into the prompt. This basically makes that process more specific (and also requires a paid account).

Though, nicely enough, you can upload a start image! So, just like Dall-E, I’m immediately hopeful that I’ll be able to generate an image based on my existing book cover. Maybe?

StarryAI prompt screen with book added

It does give me options if I want to use the start image as color, structure, depth, or scribble… let’s say I want to keep this structure and see what happens.

StarryAI results using start image structure

Ok, wow. This is the best of the four versions I was given. One was rather… lewd. Ummm… and several of the others definitely show a bias. The picture I uploaded has a fully clothed woman on the cover. The results all have a lot more skin showing—in some cases with bare legs as well. One the pose is very suggestive (and that isn’t even the lewd one!). Not too bad of a dragon head on this one though, even if the background wings/scaly tree is a bit odd.

Let’s just try to generate off of the color and see what happens.

StarryAI color from start image

This was the best of the four that were generated. Again, not… exciting? The style is rudimentary at best and it is still sticking to the structure even though I switched the input to color. Oh well, definitely disappointing if I want to keep a theme for my book covers. Let’s just try something completely new using our start prompt of a teenage girl hiding in the woods from a dragon. I’ll keep the fantasy style and see how it goes.

Starry AI results of prompt using fantasy setting.

These were interesting enough that I wanted to show you all four at once instead of just one. The final two aren’t horrible for character images, even if both are immediately assumed to be white girls who have a preference for showing off their legs by wearing short skirts. No bias there… nope. 🙄

There is something about the first one, the girl with the purple hair… and four arms… that I quite like. Her outfit is somewhere between elvish and a ranger. And purple hair with all that green—cool! But four arms… maybe if I re-rolled it a few times I’d get something that is worth keeping because right now her face is messed up and, yeah, four arms. The second one… just no again. A blonde wearing a mini-skirt… at least she almost looks to be hiding?

Let’s try using the prompt with a different style such as abstract world, since I’m sort of looking for a world setting image.

Starry AI image with abstract world prompt

You can see I added to the description as well as changed the prompt style to “abstract world.” These are… interesting. The figure is very abstract and there are no dragons except for maybe a strange toothed and scaled log in the third one from left. But, this is something and an improvement over Dall-E.

For the heck of it, let’s try again!

StarryAI abstract world style results

Ok, these are getting pretty cool! There are dragonish creatures, even if most blend into trees or vines or just seem to have extra additions to their body. And that second one from left looks more like a cobra neck? But I quite like the glowy light in them and the third one from left has some elements I’m really drawn to—even if the dragon looks to have an extra arm. The tiny, stylized figure doesn’t fit my book cover series and all of these are square since I’m not paying for pro, but this is giving me world building ideas for a different story. That’s not a bad result!

Just for the heck of it, I want to see what StarryAI can do and I’m going to try a prompt with a totally different look. hmmm… what does my story look like in Manga if I wanted to turn it into a graphic novel? 😆

StarryAI results comic style

Interesting! I really don’t mind the fourth image (on the far right). The first one isn’t horrible. Really, if you are into this style none are bad even if the horrible stereotypes of tight clothes and lots of cleavage are certainly on full display. Oh and look, all the girls are white! Sigh.

I think we’ve seen what StarryAI can do. It has some possibilities. Time to move on…

Text to Image Generator Stable Diffusion

I’ve been getting advertisements for Stable Diffusion text to image generator for weeks now. So, hey, what a perfect time to try it out?! 😆

Stable Diffusion has been out since August 2022 but it hasn’t made as big as a splash as other text to Image generators such as Midjourney. But it is 100% free and going to it’s splash screen and examples, it can generate some amazing fantasy images. So, let’s give it a try.

Stable Diffusion prompt screen

The prompt screen is a lot like Dall-E with just a simple bar to input for images. I’m not even asked to create an account or anything. Just bam, go make images. Awesome, but… well, we’ll come back to the but in a few minutes. Instead, let’s take the initiative and dive right in with our prompt.

For first try, we get this…

Stable Diffusion results first try

I see right away that Stable Diffusion is going to take my “book cover style” to heart and add actual text to the book cover. Otherwise, the images drawn are… childish? They might fit a children’s book but not something contemporary for adults—at least not something that will hold up in the market.

And I’d LOVE to show you the bottom two images. I thought I’d copied them, but… I can’t find them. AND since I didn’t have to make an account, my images aren’t saved. That is the drawback with this one. I can’t even show you that there are no options for generating new versions based on one of the images above or upscaling to a final image, because those options don’t exist in Stable Diffusion. You simply click on the picture you like—when you get one you like—and download. Or try again and lose all previous tries.

Plus, you’ll notice the images are square, which means the fake text for book names and author are cut off. If you click on the image, you don’t see more other than what is there right now (well, except for the bottom two, which I didn’t screenshot properly! 🤦🏼‍♀️)

Well, with that in mind, let’s see if we can get something better by removing the book cover style.

Those are well… interesting. Slightly horrible? The bottom right one is just WTF?

Ok, maybe it’s me and my prompt. I don’t want a drawing style but something almost real. So let’s change the prompt and try again.

Stable Diffusion text to image generator try 3

Well, there is a bit more realism. I won’t say I like the results any better. The dragons still look painted and not exactly anatomically possible except maybe the bottom left one. That one is the best out of any of the four.

But the one that gets me is the one of the bottom right. Stable Diffusion actually added a fake watermark over the girl as if this were a stock photography website and I had to buy the image. Which means this AI generator was trained using unpurchased stock photography. That actually makes me feel ill. Not only do artists not have the right to opt out of having their images used to train AI image generators to make artwork that copies their style without acknowledging or paying them, now even the originals are simply being scrapped from stock websites without compensating them. Sheesh.

Ok, I’m done here.

Text to Image Generator Adobe Firefly

I’m a graphic artist, so Adobe products are sort of like my best friends. I’m a Photoshop expert and not too far behind in Illustrator. So when I heard Adobe had created a text to image generator called Firefly that worked in Photoshop Beta, I was hyper-excited. And what is even better is that it was built using only images available in Adobe stock photography. In other words, artists and photographers who are already working with Adobe.

If that isn’t brilliant enough, they also built it with parity in mind. In other words, if you search for something like a doctor wearing a stethoscope, you’ll most likely get images of two woman and two men. Most likely all or most will also be a minority ethnicity. That is just so wonderful.

Could this be the text to image generator that I’ve been looking for?

First of all, Adobe Firefly is currently only available in Photoshop Beta, so it hasn’t been released to the public yet. But I have access to that, so not a problem. With Photoshop Beta loaded, I open up a canvas actually size to fit a book cover (yay!) and type in my prompt into the text box and hit go.

Photoshop Firefly text to image generator

As you can see, I get three images to choose from and all of them are… well… The girls look okay, even if the larger version in the image above has extra fingers. But the dragons… This is the problem with only scraping Adobe stock photography and that circles us back to the problem that there just isn’t that many good dragon images. And despite the desire to include minorities, all three girls are white.

Well, that is slightly disappointing but maybe the image results will improve before they release the full version. For now, let’s do something that none of the other text to image generators can do: build an image layer by layer.

If you haven’t used Photoshop before, the key to how it works is by having different images on different layers. That way you can move images around without affecting the ones above or below it. And you can do the same thing with the image generator!

So let’s start with adding in a forest background and a woman in the picture.

Adobe Firefly creating an image from scratch

Well, the forest turned out pretty good. I selected a misty looking one out of the three I got from my request. But you can see the girl in all three pictures isn’t right. Her face is rendered poorly. I asked for a roman style dress and ended up with three woman who might look better in a horror story rather than fantasy. I could keep playing with this and trying for a different pose. The prompt generator renders a shape based on what shape you select. I made a rather square box in the bottom right of the image and got rather square images of woman standing like boards. If I did an outline more like someone running, the result would probably be totally different.

In general though, Adobe Firefly is able to keep the lighting, tone, perspective, and focus consistent as it adds in different layers. These are aspects that are incredibly difficult and time consuming for someone doing photocompositing. So, I am still really excited about this technique… I just think it might have some more growing to do.

Still, let’s finish this up and see if we can’t get a better dragon if I highlight an area that looks like a head…

Adobe Firefly adding in a dragon's head.

Well, that is sort of what I expected based on the first image. I get a dragon’s head that looks more like a paper mache sculpture for all three options. This just won’t do for a book cover. Unfortunately, for fantasy work at least, Adobe Firefly isn’t there yet.

But I’m not out of options yet, because there are a few other things this AI can do integrated with Photoshop. For example, I can upload my book 2 cover and make changes to it and possibly even extend it.

Let’s start with extending the cover. Say that I had only purchased an ebook cover and I want to now make it a paperback wrap. Adobe Firefly has the option to extend the canvas and allow it to generate a fill based on the rest of the image. Let’s give that a try! I just really want Adobe Photoshop to work!

Regenerative fill using Adobe Firefly in Photoshop

Waiting for the AI to process my image and extend it to the white area.

Adobe regenerative fill options

You can see the AI just doesn’t know what to do with the dragon. There just aren’t enough similar creatures in its database to understand what it is and what it should look like. However, it did understand the architecture of the arch and the wall and did copy over the flames. Maybe, eventually, the AI will understand dragons too… Until then, I have seen some really cool things it can do with more realistic style covers and I have something neat for you below when we get into Midjourney.

For now, let’s try one last trick here. Let’s go back to the book 2 cover and this time just pretend that the story for book 2 has changed and we want to redo it. Maybe it is now in the forest instead of a city wall… can we change out the background?

The first step is using the auto-select for grabbing the background.

Using Adobe Firefly to change the background of an image

Now, I just need to tell the AI I want a misty forest instead.

background change in Adobe firefly Photoshop regeneration

That isn’t too bad, actually. There is some odd smooshing around her nose and I didn’t grab the hand too cleanly, but I could probably get that to work.

What about if I were to just keep the original image and wanted to change up her hair and dress. Maybe I want red hair and a dress with more style? First, I’ll select her hair and ask for red.

Using Adobe Firefly to change hair color

This one is actually pretty sweet. She’s gone from chaotic curls to straight red hair and the AI generator even filled in a bit of her shoulder and dress to replace the more smooth locks. Talk about a way of changing up stock photography into something closer to a character without spending days trying to do it!

Let’s try out her dress now…

Adobe Firefly changing a figures clothes

Wow, that is a beautiful dress. I sort of like it better than the one I had in the cover! 😆 However, for now, my challenge was to create a book 3 image, not redo book 2. So, with that, I will say Adobe Firefly isn’t there yet. At least not for fantasy covers. I still love Adobe products though and I’m really hopeful about where this one is heading.

For now, we’ll move on to what has become THE fantasy AI text to image generator and run it through it’s paces!

Text to Image Generator Midjourney

Midjourney is a text to image generator launched only in July of 2022 and already it has made tsunamis more than waves. With version 5 released in March of 2023, it could create images so realistic that they were used in misinformation campaigns on Donald Trump being arrested to false images of the Pope. This is power and powerful, which is causing a whole lot of headaches for the world.

But… I just want to make a book cover today and not get into deep fakes and false information campaigns. Why must we use these AI tools to tear apart humanity and not make it a better, prettier, and a smarter place? WHY??????

Okay, cry to the universe unleashed, let’s get into Midjourney.

For all of its power, Midjourney is currently only available on Discord. You have to be invited into Midjourney to get access (through this link), log into Discord, choose a newbie room, and type the prompt “/imagine” along with whatever you want it to create and then you are off and running.

Midjourney creates flat out amazing images. Don’t believe me? Well, let’s see how it does creating a book cover… and if it can match our existing series.

Midjourney results first try

Are you impressed yet? This is try one and just asking for the basic prompt. The face of the girl… the dragon… just wow. And all of the dragon’s are different. There is a weird word, “SAR” in the one image. Maybe Midjourney was trying to take the book cover literally. And the dragon’s legs in that one are odd. And all the girl’s remind me a little too much of Daenerys from Game of Thrones. That is the hidden bias coming out. I ask for a young woman and I am immediately get a white young woman who looks like one of the most popular women with dragons in the current period. I guess I shouldn’t be surprised.

But I also have all square images. You can actually ask Midjourney for some different sizes, but I forgot to. Well, if I choose an image say the top left one, which is officially in the first position, and ask for four new versions but with a more book cover size, let’s see what happens.

Midjourney new version with a different size request

You can see I get four new versions based on the first and they are all a better size for a book cover, BUT instead of reorienting the image, it stretched all of them. So, the proportions of face length look off. Okay, well, let’s just try this again then.

Midjourney, landscape results

I actually put the size request in backwards. I wanted to ask for 9:16, not 16:9, but this was actually a useful mistake. I could maybe be thinking of a paperback wrap. And, I can see even though ALL of these images are completely amazing, none would work as a paperback wrap where I need the main image to be focused on the far right side. The image is centered or to the left.

So, let’s fix the size request and try again!

Midjourney, asking for a book cover

This time I let it now I want a fantasy book cover and Midjourney created images that you would expect to see on a book cover, including giving us one title “Hage.” Interesting. What if I gave it a book title and author name to use? Would it be able to add those to the image? Let’s try a remix of #4 with some specific requests on name and author.

Midjourney asking for author and book title

Well, it is like Midjourney didn’t even understand the request. I just got four variations of “Hage.” So, let’s try this again to get the proper size and some better options.

Midjourney prompt results

I mixed things up and asked for an elf for just the heck of it. Again, all of these are quite interesting. I really like the color and red motes in the top left, though all would work as a book cover. And OMG, the dragons! ❤️

So, let’s say I really like the top left image… because I do. But I want to use it in a paperback wrap. It’s too small. Except… I have Adobe Firefly in Photoshop. How will it work if I want to extend this Midjourney image?

Midjourney image extended in Photoshop with Firefly

Just OMG. Adobe Firefly did an amazing job matching the color AND the lighting effect AND even added the red flecks to the part of the image it generated. It looks like one seamless picture with the tree is extended and that hazy light added. I absolutely love this image.

However… that light gold spot, though beautiful, is going to wreck havoc with any text used on the back. I need a less bright background. Fortunately, one of the three options Firefly just created for me is perfect.

Adobe Firefly extending an image for a paperback wrap

This is exactly what I’m looking for to make a paperback wrap. It fits the image seamlessly without being overwhelming for any text elements.

If I do want to play with it a little more and add something to the picture, say some magical glowing mushrooms, I can lasso an area for them on the tree and ask Adobe Firefly for some glowing mushrooms and get this…

Midjourney image extended in Photoshop with glowing mushrooms added

These do work! But… they also take away the focus from the girl’s face. So, no, I think I’m going to skip the mushrooms. Instead, Let’s really see what this image will look like as a paperback wrap!

Midjourney image used as a paperback wrap

Just YES. What a really gorgeous book cover. However, the whole point of this exercise was to make a book cover that fit with a series. I wanted to not just make something lovely but also fit a series I had going.

How does this Midjourney image hold up?

Midjourney as book 3

Well, putting the same style title and glow on it does make it blend in even if the girl is larger in proportion to the other two figures. The overwhelming blue color fits too as book 1 is red and book 2 is yellow-green. It could work, actually. Though… it would be interesting to have more action. It is book 3 of a trilogy, after all!

What if I try again and ask Midjourney for some more action oriented images?

Midjourney action based image

🤣🤣😆😆 Ok, these are hilarious. Is the dragon eating her head or just making her deaf as it roars in the first image? And for the final image in the bottom right… someone should really tell her hiding in a dragon’s mouth is not the best decision. OMG. The top right one isn’t bad, but that is a very modern outfit and really doesn’t fit the series.

But there are some possibilities here. I could play around with the prompt if I really wanted to go this way. If I found a way to add magic and action and the right outfit… as you can tell, this works but if I want a very specific cover, I might be trying a lot of prompts in Midjourney and I still might need to hire a designer to tweak the cover before I’m really happy.

But Midjourney… Wow. Just wow. The worse thing I can say is the close cropping of the images. But by using Adobe Firefly, I did manage to create a paperback wrap that works!

Hold the Press—Midjourney 5.2 Update!!

I’ve been working on this post for a couple of weeks because, as I’m sure you can imagine, generating all of these images and putting the post together was not an overnight sort of thing. And it isn’t like I could ask an AI text generator to spit it out for me. ChatGPT doesn’t even know these text to image generators exist. lol.

But I just found out that Midjourney released a new update last night that just made a few things I wrote up above obsolete. Well, at the very least not nearly as consequential. See, Midjourney will now allow you to zoom “out” and generate more of the image, much like Adobe Firefly!

Now, when I upscale a Midjourney image, I have the option to zoom out x1, x1.5, or x2!

So now, if an image is just too closely cropped for me to use as a book cover or add text to, I can simply ask Midjourney to extend the image. Like this….

Midjourney 5.2 new prompt

If I now upscale the bottom right image and ask it to zoom out by 2, then zoom out by 2, I get this series of images!

Midjourney 5.2 zooming out from image

This is a pretty amazing new feature, as each time I zoomed out, I got four new scenes to pick from. For the final scene on the right, it even had some options that included a second dragon.

The biggest thing to remember about this is the image isn’t getting any bigger. In other words, the first image where it is zoomed in is the same pixel size as the final image where she is just a small dot in the picture. This can be a huge issue when creating a book cover, especially a printed book cover that needs a very large, high resolution photo (you can learn more about that in this blog post on How to Make Your Book Cover Pop).

In some ways, using the combination of Midjourney and Adobe Firefly still works better as I can control how the image is generated and not have the size of the central figure reduced or risk having it blurry in my printed cover. Heck, the far right image still isn’t the correct aspect for a paperback wrap. BUT, I do like that I can get more of the scene from the original image into the picture, which can make up for some of the failings of Adobe Firefly.

Conclusion

There are a numerous AI text to image generators out there and more will continue to be developed. Out of all of them, Midjourney creates the most amazing fantasy style images, hands down. The problem? Well, there are several really.

  • There is more to a book cover than a great image. Want to know what else you should be considering with your next book cover? Check out this post on How to Make Your Book Cover Pop (recently updated!).
  • Most AI generators are created by scanning hundreds of thousands of images, including ones without the artist’s permission—or without acknowledging those artists much less compensating them for being part of the creation of the AI generator. Worse, artists have no way of opting out of being a part of this AI model training.
  • AI art is in the public domain. If another author/person gets a hold of the base image used in your cover, they can legally use it too.
  • I want to get into image sizes and resolution, but really, I already covered that in the How to make Your Book Cover Pop post, so just go read that!
  • If you have an image or style you want to match (like a book in a series), getting the prompt correct to get an AI to make that image can be tricky. It will take time and you might still need a book cover artist to get it the rest of the way.

AI text to image generators will continue to improve. Midjourney already amazes me and it keeps getting better with each release. Where Adobe’s Firefly goes is a big guess, but its sophistication to matching lighting, depth of field (focus), colors, and feel for what is already in your image is incredible. These tools are a game changer, but they are just tools. Writing a great book, knowing how to market it, knowing IF your cover is really the right cover for your book to attract the right readers… that is part of the author business. Don’t slack on the process just because you can fiddle around with a text to image generator and have it create something stunning. It still has to fit your book, the market, and attract the right readers.

Want to know something else these images are great for?

Content marketing. No, that isn’t just posting on social media. There is a whole lot more to content marketing than that! And if you want to learn how to really fine tune your content marketing to boost your book sales, sign up for our FREE content marketing interactive course: Skill Booster: Book Marketing Without Paid Ads. See you there!

Autumn

Autumn is a best selling indie author, conservationist, & world traveler with plans for many more adventures both real and fantastical! She is currently settled in the wilds of Maine with her small dragonish dog and husband, searching for a portal to another world.

Get Subscriber Rewards…

Or donate to the podcast one-time with Paypal!

Great Deal!

Access all of our courses for one low price

Written by: Autumn

Autumn is a best selling indie author, conservationist, & world traveler with plans for many more adventures both real and fantastical! She is currently settled in the wilds of Maine with her small dragonish dog and husband, searching for a portal to another world.

You may also like …

2 Comments

  1. Anni Zimmerman

    You did a fantastic job with this and I really appreciate all of the depth you went into while trying out the generators. I’ve also tried them and have been frustrated with all of the inaccuracies (and with getting white scantily clad women over and over again). I finally resorted to taking an image that I thought was reasonably good into Photoshop to manipulate it. I felt a little guilty about it, though, because I don’t know where the original works came from or if I was using someone’s artwork.
    Your discussion about Photoshop Beta and Midjourney was really helpful. I never thought about layering and compositing individual AI images, but this sounds super helpful in minimizing the the search for adequate stock footage, and since it has to be manipulated even farther, makes the result less like copying somebody else’s artwork.
    Thank you for doing this!

    • Autumn

      Hi, Anni! Thank you so much for the compliment. I appreciate it and am so glad you found it interesting and maybe a little inspiring.

      The one thing I didn’t really include in this was how the AI image generators are trained. When I learned that, well, it gave the perspective on copyright and image uses. I think it is worth understanding as well to have a truly nuanced perspective on this. This video: https://youtu.be/SVcsDDABEkM, even though it is a year old, is one of the best I’ve found to explain the process and what the computer model is actually doing. If you jump to around the 6:25 mark, you will get to the model training but I recommend the whole video as it is really well done and fairly unbiased.

      I hope it helps give you some greater detail and maybe helps your feelings around using the models. I know it helped me.

      I do feel for artists who are impacted by losing work to AI, but I still very much see a place for an illustrator. These models do not spit out exactly what you’d expect if you are looking for a specific image (as you’ve found out!). Instead, they are fun more for brainstorming or when you have no specific idea. If you want a character with a certain look in a specific setting… an illustrator is going to create that so much faster and better than AI. At least those are my feelings at the moment. 🙂

Pin It on Pinterest

Share This