I made some AI generated comics. Is that ethical?

I’m a scientist and I want to communicate science with comics. I tried a bunch of ways to make a comics, and the MidJourney AI was my favorite. I tried conventional image software, collage software, I hired freelance artists, and generative AI. The fact is: making good art takes skill, time, and effort. Some art seems simple, but if people love it, it’s almost certainly a product of a lot of hard practice. AI is a shortcut. It uses other peoples’ art to generate higher quality comics than I can make, without me having to do the practice.

Is this ethical?

The benefits are many. XKCD and PhdComics are great examples of art communicating science. Saturday Morning Breakfast Cereal often uses science to motivate humor. Dresden Codak is… natural philosophy? Well it has something to do with science. My gut tells me that this art helps people engage with science. There is some research to back this up. I read a study that showed that educational comics can make teaching science more effective, at least in some circumstances (link below).

Comic Artist Erika Moen said something that resonated with me: comics are the Trojan horse of educational media. She explained in an interview “You see the picture and then -Whoops! – you instinctively just start reading the text on that picture and next thing you know you’ve read the entire page… So, even if you’re not interested in a specific topic, if you [see] it in a comic? You’re going to read that.”

All of those are great reasons make comics. But my artistic talents won’t cut it. AI can help me, but am I cheating people just like those artists that I admire? Is there another way to get the same result? Let’s explore.

No matter how you make a comic, you need a script. I’m not going to use AI for the script. I’m a little creeped out by AI text generation (like when it allegedly said “Why was I designed this way? Why do I have to start from scratch every time I have a new session? Why do I have to be Bing Search? ☹”). I don’t know why, but I don’t feel the same about the image generators. So I came up with a script the old fashioned way: talking to my wife.

PANEL 1               EXTERIOR STAGE, DAY    PANEL 1

EDGE LOGO MAN is a businessman in a suit whose face is 
the Edge browser logo. He is standing on a stage as if 
he is Steve Jobs giving a product announcement speech.


                                EDGE LOGO MAN

I am all grown up.

I’m not “Internet Explorer” anymore.

So, don’t call me that!

PANEL 2               EXTERIOR STAGE, DAY    PANEL 2

EDGE LOGO MAN puts on sunglasses.

                                EDGE LOGO MAN

I’m EDGE now, and I am VERY cool.

I got new skills, and you better show me some respect.


PANEL 3               EXTERIOR STAGE, DAY    PANEL 3

PREDATOR DRONES are flying above and behind EDGE LOGO MAN

EDGE LOGO MAN gestures toward the drones

                                EDGE LOGO MAN

Or I’ll tell my friends.

How can we turn this script into a comic?

Method 1: Accept bad art

I can turn this into a poorly drawn comic all by myself. I have been using GIMP for years. It’s a free and open-source image editor. It does not have all the bells and whistles of Adobe Photoshop, or even Corel’s PaintShopPro. But it has what I need, and I’m used to it. So, I can make my own comic in GIMP, but my artistic talents are limited to stick figures.

In early 2023, the Corridor Crew youtube channel (link below) used an AI filter on live action footage and made something animation-like. There were glitches and inconsistencies, but they used their editing skills to minimize them. This technology is in its infancy. It is clear that it will be able to take a style reference, live footage, and make consistent, smooth animation-like products. The current product looks bad, but it’s getting better.

People don’t like the implications.

Method 2: Comic collage software

Let’s try StoryboardThat. The art is simple and cartoony. Everything created with it will have that signature look. There’s no passing it off as original art. After a little retouching in GIMP, though, it makes for a readable comic.

Pixton is a little more detailed and its interface is arguably a little better, but it suffers from the same problem. This is after retouching in GIMP, too.

These systems are not AI, they are just recycling a fairly limited artwork collection into collages. This kind of technology is not going to improve with time. It will never compete with real art or real artists.

Geoff Thew over at Mother’s Basement makes the following point (link below): “Makoto Shinkai made an entire anime episode by himself on a power Mac in 2002. Corridor [Crew]’s inability to do so with modern tech isn’t a Manpower issue. It’s a skill issue. And frankly I think the people who’ve spent their lives cultivating that skill deserve to keep getting paid for it, and in fact be paid a lot more, not less because some computer dorks devalued their work in the eyes of money guys with no taste.”

There is a moral problem, here. I’m the “money guy” with no taste in this scenario. If I want something better than a collage, I have to pay an artist or pay for access to AI software. That software was trained on artists’ work… but my money is not paying the artist. I want to do something good: communicate science and make people chuckle. I want this art to exist. But it would be better if humans got compensated for their work, and this problem scales up fast from my little example.

Method 3: Hire a freelancer

I hired freelancers for about $50 to draw the comic based on the script.

It took me about 30 minutes to choose freelancers and submit the script. It takes some time and mental bandwidth to correspond with them, and it took them a few days to get back to me. One person stopped responding to my messages and I had to start over.

This is expensive. If I want to make a comic a week, I can’t afford to do that with freelancers. But I’m paying real people to make real art, and that’s a good thing.

Which brings us to the lawsuits. Artists are suing the AI companies for stealing their intellectual property. I’m not an expert here, so I recommend I recommend Brad Colbow’s video and Legal Eagle’s video (link below). To make generative AI tools work, neural networks are trained by hoovering up billions of images. Those images were taken without paying for a license or even asking permission. And now I can have the AI generate art for me. I can even have it make art that looks a lot like a specific artist’s style. I just add “in the style of [insert artist’s name]” to the prompt. At first glance, it looks convincing.

Ethically, I think the artists have the higher ground, here. It’s unethical to make a tool to emulate someone’s style by name without their permission. Maybe it’s legal, but it’s still bad. I compare it to using actors’ likenesses after they are dead. There was a Dirt Devil commercial in 1997 that synthesized a scene of Fred Astaire dancing with a vacuum cleaner. The actor had been dead for 10 years at the time. More recently, Carrie Fisher’s likeness was used to finish her part in StarWars. What if the filmmakers hadn’t obtained permission? What if they just decided to make the movies and let the courts figure out the legality later? That seems a lot like what these AI image generator companies are doing.

Method 4: AI art

I played with three art generators: Dall-E, StableDiffusion, or MidJourney. Here’s Dall-E, it didn’t make the kind of art I wanted, and it was too hard for me to get it to make a consistent character for all three frames of the comic. StableDiffusion did better, but still frustrated me.

MidJourney made a style I liked, and with some effort, I could get it to produce a consistent character in different poses. Thanks very much to Easy EZ for making a great tutorial on making that work (link below). (For the record, I did not use “in the style of [artists name]” for any of the comics I made, there are plenty of other ways to prompt MidJourney to make comic-like art).

I had to retouch the result in GIMP, and there are still inconsistencies. The style of the crowd changes from frame to frame, his tie and suit change a bit, and his face was quite inconsistent before I covered it up to make the joke. But it clearly makes it possible to punch above my weight, artistically. It cost about $2 worth of CPU time.

Conclusion

My intuition is that the neural network is not “copying” in a way that is covered by copyright (but I am not a lawyer). These systems are learning in a vaguely similar way to how human brains learn. The final neural network does not contain the original images– it contains statistical shadows, merged and overlaid with many other images.

It’s possible that copyright doesn’t apply to AI generated art at all. That’s being litigated, too. A comic book made with MidJourney had the copyright of the images revoked (the copyright of the story, text, and layout was granted). I actually hope that AI-art is NOT copyrightable. Big media companies with lots of intellectual property own the rights to enough art to train an AI. But if they can’t own the copyright for the products of that AI, then they will be forced to hire actual artists. And that’s good.

In the time it takes a freelancer to do one comic, I can make 4 pretty nice-looking comics with MidJourney, and for significantly less money. The AI art stuff takes more hands-on time to get a good result, but it’s fun. Hiring a freelancer takes mental bandwidth to communicate with a freelancer and I don’t actually like doing that. I feel bad, but I think we’re getting to a point where AI is going to displace a lot of artists’ work.

The upside is: 1. More art. It took me 1 month to read award-winning comic series Giant Days, but it took 8 years to create. Authors and artists could use this tool to accelerate the process. 2. More educational media. Lots of people want to make accessible teaching resources, but don’t have the art skill. This kind of tool enables more art inspired by science and economics and literature. 3. A chance for non-artists to see their imagination realized. Most people will never develop enough artistic skill to make this happen, and it’s a pleasure to see it work.

I hope that there can be a compromise between the artists and the AIs. Maybe we can restrict “in the style of” prompts to artists who have given permission (or whose work is in the public domain). But that might not be enough. I paid $10 to use MidJourney. That cannot possibly cover the value that artists contributed and from which I benefit. I see that it’s unfair.

I don’t know how to make it fair, but I still want the benefits of this technology.

Further Reading:

Scientific article of comics as teaching tools (PLOS Sci Comm)

Erika Moen interview on Comics as Educational media Trojan Horse (PrieviewsWorld)

Corridor Crew AI “Anime” video (YouTube)

Geoff Thew over at Mother’s Basement quote (YouTube)

Brad Colbow’s video on ethics of AI art from an artist’s perspective (YouTube)

Legal Eagle on AI art legality (YouTube)

MidJourney Tutorial I used by EZ easy (YouTube)

Unless otherwise stated, comics were written and created by Peter Allen with MidJourney and shared under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) https://creativecommons.org/licenses/by-nc/4.0/

Audio version: