20 years ago, I wanted to have a web comic. But although I was pretty good at writing comics and coming up with ideas, my artistry with both the sketchbook and the process of getting those sketches into a format that could easily be uploaded to a website were sadly lacking.

Maybe AI image generation could supplement the skills I lack. I’ve been meaning to come up with a comic-style avatar for a little while, but the ones I try to make myself always come up looking a little like this:

So I wrote a description of myself into meta’s new image generator that dropped today and told it to do it in “newspaper comics style.”* It gave me this.

That may not be particularly accurate to how the mirror portrays me, but it is probably how I would draw myself if I were a character in a super-hero comic, so I’ll take it. I also like that little watermark. One of the editorial decisions I made when starting this blog is to always be transparent about what I produce using AI tools.

Here’s a couple of simple comic scripts that I can remember. Some of the references are probably a bit dated, like when my friend Jeff suggested I draw one about Peter Jackson’s lord of the rings movies. I have long since lost the sketchbook, but each panel was just a shot of a guy sitting in the movie theater and the title of the comic is “The lord of the rings experience.” In the first panel, captioned “The Fellowship of the ring.” The guy looks super happy and excited with an awestruck, joyful grin on his face. In the second panel, titled “The Two Towers” he has his head tilted to the side and just looks puzzled. In the final panel, titled “The Return of the King” he’s looking at his watch. I added the prompt ‘3 panel comic:’ to the front and put those last two sentences into the image generator. I must admit, I’m a bit disappointed.

It generates alternatives, and this was the only one that wasn’t just lord of the rings characters sitting around a fire.

As with all new tools, I’m going to have to spend a bit of time learning how to give it prompts that lead to exactly what I want.

Let’s try one more comic I remember writing back in the early Onesies **. Sadly, my original is in the same missing sketchbook as the Lord of the Rings one. For some reason, the idea of an AI attempting to deal with all of the absurdism in this one really makes me giggle.

4-Panel Comic. Panel one: Man 1 with red hair, green eyes, glasses and goatee bursts into room and exclaims: “How can you two be sitting around casually eating pokemon cards?! Don’t you know what day this is?”

Panel two: Guy 2 checks his phone “Friday the Thirteenth?” Guy three spits out a chunk of pokemon card and say “It is, oh cool. I always have good luck on those days.”

Panel three, Guy 1 explains “Yes, normally Friday the Thirteenth just weirdly shifts luck around, but this one also falls… On a Full Moon.” *Thunderclap*

Panel Four: Guy two, stricken: “My god! Werewolves.” Guy One: “LUCKY Werewolves.” Guy Three: “I’ll get the silver bullets.”

Let’s see what I get by lazily using each of those paragraphs to get a panel since I doubt all of that will fit.

Oh boy. Yeah. That confused the heck out of it.

Prompt for panel one: Man 1 with red hair, green eyes, glasses and goatee bursts into room and exclaims: “How can you two be sitting around casually eating pokemon cards?! Don’t you know what day this is?”

I had a feeling that saying something strange like eating a thing that is not food would cause it some trouble. At least I didn’t get any gory images of people eating pokemon.

Prompt for panel 2: Guy 2 checks his phone “Friday the Thirteenth?” Guy three spits out a chunk of pokemon card and say “It is, oh cool. I always have good luck on those days.”

It’s definitely getting the pokemon theme, but this is a long way from how it looks in my imagination.

Prompt for panel 3: Guy 1 explains “Yes, normally Friday the Thirteenth just weirdly shifts luck around, but this one also falls… On a Full Moon.” *Thunderclap*

Note to self: It is looking for key parts of the image like the moon and a guy explaining something. I think the dialog is confusing it.

Oh well, I might as well finish out this experiment. Prompt for panel four. Guy two, stricken: “My god! Werewolves.” Guy One: “LUCKY Werewolves.” Guy Three: “I’ll get the silver bullets.”

This one seemed to take longer than the others. And the results were kinda disappointing.

Lesson: Just like I had to learn with Chatbots, the AI is not that smart. If I want the results to match my expectations, I’m going to have to be really specific. I’ve already started a google doc to just keep all of the very specified prompts I use when I need to generate a multiple choice test, for instance. I’ve also had to add a few things that I will need to say as follow ups to refine the difficulty of questions and get it to not make every answer be C.

Before I go, I want to see if I can get it to show someone actually eating pokemon cards. I’ll spare you from anything that is too disturbing.

[About 20 minutes later.]

The good news is: I didn’t get anything too disturbing, but I had to adjust my prompt a lot before I got anything close. Here’s what I landed on.

Prompt “Two guys are casually sitting on the couch snacking on pokemon cards. They are both eating the cards as if they are potato chips even though cards aren’t food.”

It really really wanted to show people holding a hand of cards.

That’s enough silliness for now.

What did we learn today: AI image generators can be used when you don’t care too much about the result, but to get something precisely like what you want, you probably need to use a bunch of specific terms of art to specify the orientation of the objects, and get extremely specific about your descriptions of the subjects of the photo.

Also, definitely avoid using text that isn’t directly related to how you want the result to LOOK. I’ve managed to get it to put in word bubbles, but I’ll have to edit the text into them myself.

Also, as with all AI, giving it context by using the word ‘style’ to tell it what type of thing your looking for goes a long way.

I tried and tried to get it to draw me as a two-dimensional cartoon, but it couldn’t seem to do that and I’m ready to give up. Honestly, the first one is so over-the-top that I giggle every time I see it. Also, the green eyes were frequently glowing or emitting lasers for some reason.

Here’s the prompt I eventually got to and the image it generated before I gave up.

Prompt: ‘Man with auburn hair that’s a little blond, green eyes, glasses, a goatee and a small smile in the style of simple, line-art cartoons with minimal details, Like XKCD but with eyes.”

Footnotes below.

*The exact wording of the prompt was “Man with auburn hair, glasses and a goatee in newspaper comics style.” what it returned was similar, but way thinner than me and had the wrong eye color so I added. “Add green eyes and rounder cheeks.”

**My term for the period from 2000 to 2009. I need to make a page I can link to with editorial policies like this one that I need to explain; these footnotes are getting out of hand.