So, everyone is having a lot of fun with the new ChatGPT image generation — mainly making Studio Ghibli versions of images. But another thing it’s very good at is text, and it can put paragraphs of text in an image without making a typo.
So, that brings up the question: Can it make comics?
I decided to test that out by just giving it a premise: I told it to make a comic about Nikola Tesla and a duck teaming up to solve crime. Here’s what it made:
Okay. Well, it has a consistent Tesla and duck, but it doesn’t really have a plot or any humor to it. So I brainstormed with AI on how it could be funnier. It suggested the criminal could be revealed as Edison, which isn’t a bad idea. Still, that wasn’t really funny all by itself. I ended up telling it exactly what dialog to put into the last panel to make it funny.
So, interesting experiment. Suddenly, I had some ideas for a few more comics, so I decided to see if I could make some consistent comics for it. I decided to go with the name “Slop!” as a reference to AI slop, as I really am doing this in the lazy creative way, me being a failed comic artist (I had ambitions as a kid but never stuck with it) and being fine with whatever mediocre style AI comes up with.
So, I gave ChatGPT a picture of me and told it to make a comic of me and AI, but this time I fed it all the dialog. It really had trouble with multiple dialog balloons and making sure they pointed to the right character, so I ended up having to cut down the dialog (again, Slop! is a good name for this having to compromise the original vision), but this turned out pretty good.
Next, I wanted a longer one, so I had it do this in two parts. Again, it took some tries to get the parts right (though this helped by having only one character speak per panel). And sometimes I’d have to manually edit the little thingy off the dialog bubble (what’s the word for that? good question for AI) to point at the right character, but pretty good, though it didn’t keep the person super consistent (I gave it the previous comic as a reference).
Then I had an idea for one with different characters. It kept fighting me on this one because it didn’t want violence or a comic involving Batman though you don’t actually see either. I had to manually darken the second panel to make it look like there was a shadow over him.
Now I tried doing it one panel at a time and concatenating all the panels after (BTW, ChatGPT 4o is good at basic image manipulation writing and running Python scripts on the fly, so I just had to hand it all 8 panels to put it together). This still took a lot of tries per some of the panels because it has trouble fitting in the dialog and making sure they point at the right character. Also, I became fatter for some reason.
And here is one last one, and it solved some of its own problems by only having one character in the panels at a time. Took a few tries, but it eventually did this in one shot though I had to flip one panel because the computer was on the left.
Anyway, just an experiment. Maybe I’ll make some more comics if I have some ideas. It’s a bit frustrating to work with, as the main problem of ChatGPT image generation is that if you tell it to change one part, it has to redraw the whole thing (it can’t yet do what’s called “inpainting,” where it just changes — injects noise into — one highlighted section). Still, this is probably a lot faster than drawing the comics myself. I know a lot of people hate AI for one reason or another, but the future is going to belong to those who learn how to use it — especially those who can learn to use it to get a specific vision without compromise.
It seems like the more you use AI, the younger it makes you. That duck got young fast, and in your first comic your hair has a bit of gray to it but not in the rest of your comics. Be sure not to use it too much or you'll revert to your baby stage. And then maybe even cease to exist!
Tesla & Duck is like a Michael Kupperman comic.