I Beat AI Today
How I created something that AI couldn't.
AI can’t think. It also can’t be creative. This is where real artists live.
I recently wrote an article about how AI broke me. The new ChatGPT image generator can create comics that are good enough for most people, and if all I can do is create something that AI can create, then there’s no reason to not use it to generate images.
But today I beat AI.
This morning while dozing in bed I came up with a creative way to illustrate the phrase “conversation is a game” for a recent project. I thought, why not make it an actual board game? With students walking on it as they go through the halls? I’ll do it as a birds-eye view and show other kids in the halls.
I knew this would be a hard image to draw, but I was sold on the concept and set to work. Here’s the draft of what I came up with:
This is a complicated two page spread, but my creative monkey mind insisted not only that I could do it, but that it helped communicate the point of the narration.
There was so much going on here I was pretty sure AI couldn’t duplicate it. Let’s find out!
Here’s the first instructions I gave it: I need a two page comic book spread with two girls walking down the hall talking. They should appear in the spread five times to show the passage of time. They are walking along the hall on what looks like a gameboard path. There should be various kids in conversation around them.
It generated this:
The dialogue is passable. It doesn’t really sound like teens, but it’s not awful.
But the backgrounds lack any kind of narrative sense. None of the panels seem connected to the one before because the background keeps randomly changing. There’s some weird perspective issues where the two girls are walking on the same level as the game board, but the characters on the periphery are below the board (and the horizon line). All of sudden there’s a lunch table in the middle of the hallway, and a lunch tray that appears and disappears. The expressions are the same throughout.
If someone uses this image, they are clearly just looking for something that will do the trick. Or they might have the same idea as I did without the knowledge or drive to pull it off. But this is the kind of problem that makes life as a creator so fun.
Should I give AI another try? Here are the clarifications: Can you do it from an overhead view? And make it one large panel? (Honestly, I had trouble telling AI what I wanted it to do because the concept was hard to describe.)
It’s getting closer to what I was up to, but still pretty terrible. The poses and expressions are all the same and the path doesn’t follow any logic (nor are the girls following the path). The same issues with perspective are there, as are the random school activities (now the lunch table is in the middle of the hall!) And lots of speech balloon tags!
Could AI pull of something like I did? Maybe eventually. But right now it can’t come up with a complex scene that makes logical sense. It can’t think something through as a whole, so what you get is a mishmash of images that suggest that AI got pushed to the wall and had to find some way to get out of it.
So as I’ve said before, comics creators need to lean into what AI can’t do. It can’t generate complex images. It can’t problem solve and can’t come up with imaginative solutions to problems. This is the part of the craft that humans excel at, and will attract people that want to work with us and value our expertise.
I also think it’s important to not draw like AI does. Unfortunately, if you glance at any middle grade graphic novel section in the bookstore, you’ll see a lot of character designs that look like this. We need to create a style that stands out and is appealing in it’s own right, that makes people say “I want to engage with this.”
For example, one of my favorite comic artists is Richard Thompson. His style is so idiosyncratic it’s hard to believe AI will be able to spit out a Richard Thompson strip anytime soon.
So lean in to what make you unique!
And if you like what I’m doing, please subscribe!