My AI-created comic won’t be putting any artists out of work just yet

Hey, there are no scientists in that one! And my antiheroine Scarlet Shadow certainly doesn’t look like a cross between a human and a demon ... After just a day of using artificial intelligence (AI) to create a comic, I throw in the towel. It had all started out so promisingly too.

Chatbots and intelligent image generators are both fascinating and unnerving at once. They appear to be surprisingly creative, seemingly producing content out of thin air. Seemingly so, because the information or images they produce come from an existing stock. What’s more, issues such as the crediting of sources and image rights continue to raise plenty of questions. My colleagues Michelle Brändle and Coya Vallejo have already explored some of them. For my project, I focused on feasibility, not legal aspects or the question of what art is.

I need a character

I’m a huge comic fiend. Alas, I lack the talent and motivation to draw one of my own. But maybe new tools such as Midjourney and ChatGPT could help me make my dream come true after all. Midjourney is a program that uses artificial intelligence to create images from word input. The tool is currently in beta stage and is operated via the chat software Discord. Typing the command «/imagine» into the chat box followed by any word allows me to generate images.

First, though, I need a story. And every story needs a main character. This task falls to the chatbot ChatGPT. Given that the documentary «Mike Mignola: Drawing Monsters» has rekindled my love of «Hellboy», I use the red demon as inspiration. I type in:

I need a character for a comic book. It should be part human, part monster and should be based on Mike Mignola’s Hellboy

In response, I get Scarlet Shadow – a cross between a human and a demon, with scarlet fur, long, pointy ears and a tail. Although summoned by an evil sorcerer to serve as his weapon, she has a mind of her own; deciding instead to turn against him and join the good guys. After being found by a team of paranormal investigators, she helps them in their battle against supernatural threats.

So far, so Hellboy. At least when it comes to the superpowers and ability to shapeshift, there are a couple of interesting deviations. It’s perfectly fine as a starting point. Let’s move on to the story. Here’s my attempt at creating one:

Tell me a story including Scarlet Shadow that can be told in a 15-20 page comic. It should be set in the 80s, contain supernatural and science-fiction elements, and take place in Switzerland.

I change the suggested title – «The Alpine Eternals» – to «The Eternals». The rest of the story is gloriously trashy comic book silliness. It revolves around a group of scientists investigating a supernatural phenomenon in the Swiss Alps. Whilst there, the crew is attacked and kidnapped by unknown actors. On patrol nearby, Scarlet Shadow hears the commotion and resolves to get to the bottom of it. In doing so, she discovers «The Eternals», a secret organisation whose members believe they’re the descendants of an ancient Alpine civilisation. Not only that, but they’re embroiled in a conflict with aliens.

With some further questioning, ChatGPT gives me details where needed on exactly how Scarlet Shadow is able to free the scientists and, alongside the Eternals, defeat the aliens.

Pictures, please!

The information I’ve gathered should be more than enough to create the first images. The Midjourney image generator allows me to be really experimental. I start with a keyword-style description of the Scarlet Shadow character suggested by ChatGPT.

imagine/ woman with scarlet fur instead of skin, she is a mixture of human and demon, the woman has long, pointed ears and a long tail. She also has powerful claws and teeth as weapons

When it comes to style, I rely on my own preferences. Here’s what I type in: 80s retro sci-fi future art style, scarlet, graphic novel, --ar 3:2. That last part is a parameter for the image format.

It took at least half an hour for this picture to emerge, with one task taking roughly 1-2 minutes. Consequently, I took out a subscription for 30 dollars, which allows me to set up to three quick tasks at once. During that half an hour, I experimented with various words and word orders. It’s not quite clear to me which words Midjourney prioritises and how. What it doesn’t seem to want to get right is Scarlet Shadow’s scarlet fur. The best drawing Midjourney can do is a woman with a furry scarf around her neck or a cat on her shoulder. Oh well. I’m certainly satisfied with the cover image.

In the same style, please

The challenge is now to create an entire comic book in the same style, with a main character that looks the same throughout. A US American designer, who used the same tools to make and sell a children’s book on Amazon, had the same problem. His characters don’t look identical, and that’s with keeping things to one picture per page for the most part. My goal is to create a real comic with multiple panels per page.

To continue the style as much as possible, Midjourney allows for the command «seed» followed by the seed number of an image you’ve created. I want the cover to be my anchor point, which gives me a parameter of --seed 4187562391.

I add it to each set of keywords for the images that follow:

80s retro sci-fi future art style, scarlet, graphic novel, --ar 3:2 --seed 4187562391

Now I have a heroine, a story and a style. Next on the agenda is to ask ChatGPT what the first page should look like. The chatbot comes up with templates, including the one for the first panel – a panorama of the Alps under a threatening sky. Next to that is a smaller panel depicting a group of scientists with equipment on their backs, ready to set off. In another small panel, Scarlet Shadow is on patrol in her human form. So far, so good. As I’m transferring this over to Midjourney, however, the problems kick in. I manage to get the panorama in pretty much the same style as the cover image.

The panel with the scientists takes considerably more attempts before I’m happy with the perspectives and how the figures look.

Deviating slightly from ChatGPT’s suggestions, I then decide I want another picture of the scientists. A close-up, where I’ll be able to create a speech bubble. In addition to the drawing style, the appearance and skin tone of the figures now has to be right, too. Generating a speech bubble doesn’t work at all, but I could insert one manually afterwards. Numerous text combinations later, I get a suitable image with this command:

swiss scientists, with futuristic equipment, hiking up the mountain, close up of two scientists, sky is dark, cold, eerie, Scary 80s retro sci-fi future art style, scarlet, graphic novel, --ar 3:2 --seed 4187562391

Later on, I drop the «Swiss» from «scientist» because it never draws the Swiss cross I’d been hoping to get on their suits. Creating the next panel has me improvising, too. According to the story, the scientists were investigating mysterious phenomena before getting abducted. Consequently, I want a picture of them discovering something pulsing with supernatural light. It takes almost two hours before I finally manage to get the scientists facing backwards, looking roughly the same as in the picture before, staring at a glowing object.

At the next panel, I cut the project short. My goal is to have Scarlet Shadow at a medium distance away, secretly watching the scientists as they approach the UFO. No matter what I try, I simply can’t get Scarlet Shadow to look like she does on the cover – let alone get her to watch the scientists from a distance. They’ve totally disappeared. Who knows? Maybe they’ve been abducted by aliens.

Occasionally, the style also changes to 1950s, Swiss, black-and-white postcard photography with a red UFO in the sky. I give up.

Artists don’t just fall from the sky (or The Cloud)

My experiment quickly revealed both the possibilities and the limitations of AI as an artist’s tool. I managed to create awesome images – some of which went straight into my collection of background images. When it comes to the details, however, many of them don’t cut the mustard. There are half-finished ears, light shining through where it’s not supposed to be, and numerous other irregularities.

However, the main problem in generating the images is the lack of consistency. I didn’t manage to keep the drawing style going. Especially when it came to keeping a figure’s appearance the same or customising a scene. Typing in too much information seems to overwhelm the Midjourney image generator.

What I can hardly fault, however, is ChatGPT. Sure, the story it drafted certainly won’t win any Eisner Awards, but it does provide a useful template for an entertaining comic book story.

We’ll be dealing with the art and plagiarism implications of artificial intelligence for a long time to come. As a content creator myself, I find the technology (despite my failed attempt at using it) extremely exciting, although still concerning.