A first-hand exploration of AI art and its viability in comics.
Well, those Silicon Valley eggheads finally went and did it. They made artists obsolete, kaput, completely and utterly unnecessary. AI art has come to replace us all. Or has it?
The visual arts have long evolved hand-in-hand with technological advancements. The printing press, the camera, and the personal computer are all inventions artists have embraced and used for new avenues of creative expression. Can AI art provide those same opportunities? Or are we all doomed to bow down to a machine? Those are the questions I asked myself as I began to investigate several AI art applications to see just what I could get them to produce. What follows are my personal thoughts on AI art after spending hours prompting, experimenting, studying, and, yes, even incorporating some of it into my own comics and zines.
What is AI Art?
While such technology has been developed for decades and taken many forms, recently easy-to-use AI art applications have become more accessible to the general public. They are text-to-image models, meaning you type in a prompt, and, once you hit “return,” an image, or collection of images are produced. In addition to the described subject matter, a user can include a variety of details to flush out their images, such as descriptive adjectives, illustrative or photographic styles, lighting direction, aspect ratio, scenery, periods in time, and even include artists for it to mimic. That last option is one of the main points of contention that has AI art in hot water.
Currently, a trio of artists have initiated a class action lawsuit against businesses Stability AI, Deviantart, and Midjourney. They argue that these companies used millions, perhaps even billions, of images obtained from the web to train their systems without compensating nor asking permission from their original human artists. Furthermore, these ill-gotten images can be used to replicate and mimic a particular artist’s work, thus costing them jobs and revenue. Stock image giant Getty Images has also filed its own lawsuit, claiming these applications have infringed on their copyrights as well.
On the other hand, the companies in question claim their actions fall under fair use. They
defend AI art as being more than mere engines that collage together images
on command. On the contrary, AI art runs on algorithms and builds completely original imagery from scratch based on mathematical representations of the data they have collected. Despite this, it should be noted that many AI art applications have implemented “optout” options for concerned artists that promise to restrict images from being implemented in future AI training. Critics say those measures are still not enough.
Whatever the outcome, it will take years to reach an answer to these legal and ethical questions. To make matters even more complicated, copyright and fair use laws differ from country to country, so AI art could become more restricted in some parts of the world over others. If you’re curious, you can read more about the lawsuit at stablediffusionlitigation.com.
For those of you concerned, aside from a handful of attempts to test AI art’s ability to replicate the work of some comic book greats (like the facsimile of Jack Kirby’s Thor below), I mostly avoided using any specific artists in my prompts. While I have no doubt AI art can competently replicate the work of many artists, the results I received when trying this were less than stellar.
Of the AI art applications I experimented with, Midjourney proved to be head and shoulders above the rest. It consistently gave me polished-looking results and will be the main program of interest for this article. Midjourney is a pay site, however, with the lowest tier being $10, so if you are interested in dipping your toe into AI art, I’d recommend starting with a free site, like Hugging Face Stable Diffusion (huggingface.co/spaces/stabilityai/stable-diffusion). You will receive poorer-quality imagery based on your prompts, but it’s a good initial testing ground.
What AI Art is good at
Simply put, AI art is capable of producing high-quality, professional-looking imagery with nothing more than a user’s imagination and an internet connection. It is the very definition of instant gratification, taking only minutes to spit out illustrations that would take the human hand hours of labor and practice. AI art has already been utilized in magazines such as Vogue Italia, Photo, and Cosmopolitan, though one could argue this was done more as entertaining a new fad than using AI for its artistic merit. Judges at the Colorado State Fair even declared a piece mainly comprised of AI art a first-place winner in an art competition last year.
Upon first using Midjourney, I was immediately impressed with the slick, detailed work it produced. Its default style appears to be a painterly photo-illustrated look that offers an aesthetic that’s sure to satisfy the vast majority of users. Celebrity likenesses and portraits are an especial strength, as well as fashion and clothing. Perhaps not so strangely, I also noticed all AI art applications I tried were very good at creating interesting robots. Color usage is another strong point that they were all good at across the board.
What I personally like the most about AI art is seeing how the programs interpret my odd prompts. Subjects I have input include a claymation insect beauty pageant, an avant-garde editorial layout for a lunchmeat magazine, elaborate geometric 1950s Jello molds, a vintage Cthulhu Sears catalog ad, and cereal box art created by famous artist Joan Miró (see below for some of the examples). Results were mixed, but it was fun to see what possibilities the machine would spit out.
Beyond just typing in prompts, users can also upload their own images to use as reference. These images can be stylized or even combined using Midjourney’s “blend” command. Both of these features offer artists opportunities to see their own work in a new light and can provide them with inspiration or new possibilities to explore they hadn’t thought of. I like to think of this as a glorified magic sketchbook, able to provoke new perspectives.
Overall, I found if one has no preconceived notions, the AI results were usually at least interesting to look at. It was only when I had something specific in mind, more complex, that the programs became frustrating.
The stuff AI Art isn’t that great at
Let’s start with an obvious one. The most well-known failing of AI art is its often inability to create believable human anatomy. I’ve witnessed all manner of grotesque monstrosities with arms and legs contorted, multiplied, or merged in horrifying ways. These bold and strange freak shows were actually what first attracted me to experiment with AI art. There’s nothing inherently wrong about enjoying these images for what they are, but if your ultimate goal is to illustrate a particular scene or idea, these oddities distract from the overall message. Hands especially have become a meme at this point for the downright mind-blowing mutations the AI generates. In all my attempts to have it create a believable handshake, none came close. In general, it seems AI art does well at creating stoic humans cropped just above the waist that passively look straight ahead for the camera. Getting a figure in any action pose, such as running, grabbing, holding an object, swimming, etc, is a gamble, with nearly all of my attempts deemed unsatisfactory.
The glossy, digital veneer most Midjourney images have that I mentioned earlier probably appeals to most people, but the more I used it, the more stale it became. For me, there’s an unnatural quality about it that’s inhuman in how ideal it is. Even as I tried to get it to express other, more photographic and realistic styles, the plastic artificialness would consistently creep back in.
On the more practical side, resolution can be another issue for anyone who would like their work to be printed. Since these are being produced on the web, the files are typically 72 dpi (dots per inch), whereas a good target for print quality art is 300 dpi. Midjourney does offer an “upscale” option that helps alleviate the problem by increasing the file sizes to 1024 x 1024 pixels (roughly 3.4” x 3.4” at 300 dpi). According to one source, Midjourney can scale images up to a maximum of 1792×1024 pixels (or about the equivalent of 8” x 10”), but I have yet to progress to that level. There are also methods using third-party programs Photoshop and Lightroom that promise to get even bigger print-ready images with minimal
pixelation.
Beyond these creative and technical shortcomings, ownership of AI imagery is also another drawback.
While many legal matters have yet to be addressed, a situation recently played out that might set a precedent for the future. According to an article published by theverge.com in February of this year, the U.S. Copyright Office denied partial copyright to a comic illustrated with Midjourney entitled Zarya of the Dawn, stating it included “non-human authorship.” The writer, Kristina Kashtanova, claims her direction of the AI algorithm and arrangement of the images into sequential form should constitute the work as her intellectual property. Her lawyer adds, “AI-assisted art is going to need to be treated like photography. It is just a matter of time.” The Copyright Office also ruled against the AI itself having the authority to hold copyrights on work it generates, comparing it to the famous “Monkey Selfie Dispute,” where it was deemed non-human actors are not legally entitled to copyright claims. These rulings appear to not necessarily preclude work with an element of AI art from being copyrightable, just those without the appropriate input of a human hand.
The AI applications themselves also have their own sets of rules when it comes to this issue. I’m unfamiliar with the terms and conditions of all platforms, but in the case of Midjourney it is stipulated that you can use imagery generated for commercial use (provided you follow certain standard guidelines), yet you have no exclusive ownership over the images you create with their service. Additionally, since they operate using public Discord servers, any user can snag the work generated by another user for their own purposes. The “sharable” nature of Midjourney is actually a feature they advertise, as you can search through reams of user-generated art once you make an account.
Can AI Art be used to make comics?
While there’s no doubt AI art can create pretty pictures en masse, there’s more to making comics than that. As I delved deeper into testing Midjourney’s capabilities, I found getting it to create images that tell a story to be very difficult. Foremost, creating consistent characters in a variety of poses was an overall failure. Even when uploading reference images, Midjourney would add subtle changes, such as altering the hairstyle or clothing. Sometimes it would do it slightly. Other times it would completely change major features in an undesired direction. Capturing the same character from different camera angles was another challenge I stumbled into. Straight ahead, profile, and three-quarters shots worked out okay, but Midjourney would usually only generate these figures cropped at the waist. Getting it to give me a variety of angles showing a full-bodied character never happened, no matter how many alternate ways I input the prompts. You can see some of what I’m talking about in the images to the right.
I also attempted to get the same face to express a range of emotions from happiness, to rage, to surprise, to sadness. Midjourney can nail the smiling part. The others may or may not come through. Many times, it generated faces that appeared smushed, contorted, and confused. Perhaps those are fitting representations from a machine that has never known emotions.
Moving on, I tried to put some characters into an action scene, especially those you might see in a comic. Throwing a punch, shooting a gun, someone reeling in pain, and even something I thought would be simple, like a woman slapping another woman, took a slew of tries without many results I was happy with. Introducing more than one figure into the mix, like two people in a struggle, never quite seemed to work out either. To make matters even more difficult, sometimes the wording of these prompts were on the program’s list of banned words. More than once, a warning message popped up that my entire account could get suspended if I continued, forcing me to carefully reword my prompts. The reasoning behind such a measure is obvious, as I’m sure bad actors can think up all sorts of sick fantasies to illustrate with this technology. It’s a safeguard I don’t disagree with, but it solidifies my experiences that getting it to generate a solid PG-13 rated fight scene
is tough.
Getting Midjourney to create anything with words in it is also nigh impossible. You might manage to get something short, like “Stop” or “Go.” Anything else will appear as an alien language or even be simplified down to marks that resemble a bar code. For now, at least, human letterers are irreplaceable by the machine.
With my results less than stellar, I decided to look at some other creatives that have managed to make comics using AI art. After a quick Google search, I found campfirenyc.com and downloaded some free PDFs of a couple of their most recent AI-produced comics, The Bestiary Chronicles. Written by Steve Coulson with art generated by Midjourney, this multi-issue comic series has great colors with pleasing illustrative styles. Coulson managed to keep the overall look cohesive throughout. More importantly, I found he had unlocked some secrets that allowed him to get more consistent characters in his work. Based on these qualities, I felt these comics were fitting benchmarks to examine.
As I looked closer, however, I immediately spotted that he encountered some of the same problems I did. For instance in issue #3, “The Lesson,” the main character is a blond woman wearing a blue jacket. She remains identifiable due to these traits, but her features, like her hair and clothing, varied on almost every page. There is also a lack of variety of poses, with zero full-body shots of her. She never performed any action of any sort, more or less just acting as a mouthpiece explaining a story to a class of children, and never once did Coulson show her hands in 52 pages. The storytelling reads more like a collection of images stuck in a sequence rather than an actual comic.
The characters are much better in issue #4, “The Letter Home,” with more variety of poses, and even some good hands are shown. One impressive image Coulson did get was a soldier believably holding a baby, something that gave me strange anatomical mutations when I tried it. On the downside, the issue again misses those vital storytelling elements that make an exciting comic. It read like separate images stacked in order and lacked an interesting flow. The action is still missing, especially in the war scenes. Soldiers mainly stand statically still with their backs to the camera. Giant kaiju monsters are attacking them in the jungle, and no one is running in terror, no one is trying to escape, and no one is even firing their weapon.
Do these shortcomings even matter to most publishers, though? In an article published by cnet.com, Coulson comments, “By the new year (meaning 2023), even the trained eye probably won’t be able to perceive an AI-generation from any other. It’s exciting and terrifying at the same time. But you can’t put the genie back in the bottle, so we’re embracing the future as fast as we can.”
Putting AI to the test
With all of this being said, I decided to see just what kind of comic I could make using nothing but AI. Not only would I use Midjourney, the writing, too, would be AI-generated with Chat GPT handling the script. You can check out the results on the opposite page and
judge its aesthetic and entertainment value for yourself. Personally, I found the experience to be extraordinarily frustrating.
Aside from the disembodied hand in panel 2, everything started off well enough. By the time I reached panel 4, however, I was already forced to ignore details in the script, such as Dr. Sandy wearing sunglasses. When prompted, Midjourney would put sunglasses on all the
characters, no matter how many ways I phrased it.
In panel 5, Dr. Sandy was to point at the sneezing man. Midjourney instead initially gave me a serious longhaired man that resembled David Carradine for some reason. No pointing anywhere to be found. I eventually had to compromise and put the two doctor characters
off-panel to get a semi-believable sneeze.
The troubles only continued into panel 6. Chat GPT’s script called for a person acting like a seagull, but I had no idea how to word that to Midjourney. Instead, I thought the joke might be funnier and simpler to illustrate with a real seagull. Midjourney still gave me weird results. Doves instead of seagulls, seagulls with multiple heads, seagulls with three legs, and seagulls standing in midair instead of flying were all images produced by the AI.
Panel 8 was nearly my breaking point. It made the seagull fiasco feel like a walk in the park. In the script, Chat GPT directed: [Dr. Sandy and Missy approach a beach umbrella where Dr. Waves, a doctor in swim trunks and a Hawaiian shirt, treats a woman with a snorkel.]
This is when I found out Midjourney has no idea what a snorkel is. Prompt after prompt, it just got worse. When I gave up on a believable snorkel, it became apparent Midjourney wanted the beach umbrella to be the star of the show. Then there was an odd cameo from a Rosie O’Donnell lookalike. After that, I think Midjourney got fed up as well and started giving me weird sci-fi art. After spending about 45 minutes on it, I picked the only usable image and declared it “good enough.” See below for some of the rejects that had my blood boiling.
The final panel was another head-scratcher, as the script called for Dr. Waves to reappear with the two lady doctors. For all my efforts, Midjourney was determined to make him the star of the show. It would not put him in the background so Missy could get the final word. As you can see, my solution was just to omit him altogether. Another funny tidbit was the AI attempting to give everyone mustaches to match Dr. Waves’ look.
Overall, I had Midjourney produce about 350 images total, of which I narrowed those down to the twelve that made the final cut of Beach Doctors. Based on this exercise, I do not feel AI art demonstrated the ability to be a serviceable comic-making tool on its own.
Final Thoughts
As I was preparing for this article, I was reminded of a conversation I had with my roommate over 20 years ago when I was a freshman in college. He was excited about the release of Final Fantasy: Spirits Within, a computer animated movie based in part on the video game. While the film seems to be forgotten by the public today, at the time, critics and fans praised it for the photorealism of the characters. Remember, this was in 2001 and CGI was still new and exciting.
My roommate proudly proclaimed that this film was so brilliant, it was the harbinger of the end of human actors. They were now all replaceable by technology. Never a fan of my roommate’s dumb opinions, I disagreed and pointed out he was discounting the contributions of the voice actors in the film, as well as others who were no doubt used as reference by the animators. Additionally, I argued audiences will never want to see fake people all the time, especially those who enjoy the theater and other live performances. Upon reflection, that kind of sums up how I feel about AI art.
Like all great technological advances, AI art is here to stay and will keep improving, further blurring the line between human and machine. That future may feel foreboding to some, certainly to myself and the other artists among us. We naturally worry about the added competition from a program that can do what we do in an instant, negating the skills we have put so much time and effort into honing. In the long run, though, I hardly feel AI art is a suitable usurper for the nuances and expressiveness of human art, especially when it comes to storytelling. Like all creatives, we will learn to adapt. Remember that CGI has not replaced the actor, the computer has not replaced the designer, and the camera has not replaced the painter.
Following comic was created with CHAT GPT & MIDJOURNEY (with assistance from Steve Steiner)
PROMPT: Please write a 1 page comic script entitled Beach Doctors about the doctors that provide lifesaving medical procedures on patients that fall ill at the beach. It stars Missy Marie Higgins. She comes from a small town and is having a hard time adjusting to life on the bustling beach. It’s her first day on the job and she’s anxious, but excited. The other doctors on the beach will need to show her the ropes so she can succeed. The style should be absurd, humorous, and campy. Full of jokes.
