Cartooning with AI: my part of the elephant

Some mad science cartoon experiments for you...

😁Feel free to scroll down to see some examples if you don’t have time to read right now.

I’ve always loved the parable about the blind people describing an elephant.

You know, where one person feels a huge leg and describes the elephant like a tree, another feels the tusks and describes it like a giant sword, the person on the tail might describe it like a rope and so on.

That’s how I feel about AI image generators right now.

A lot of people are blown away, thinking, “now I can draw! This is amazing! I didn’t even need to learn how!”

(For now let’s set aside the issues of copyright and lack of requisite artist-suffering.)

To this guy feeling the elephant, using AI feels like a step backwards for my artistic abilities.

Sure, I can generate amazing stuff with a couple keystrokes, but it doesn’t look like my stuff.

Sparkly three-dimensional superhero art? No problem!

Cartoon characters that look hand-drawn in my style? That’s a lot trickier.

For the most part I’ve been frustrated because I see an image in my head that I could draw the old-fashioned way, but I can’t make the robot do what I want it to do.

Yet.

While I’m frustrated now, I already see the huge time-saving potential for fast turnaround animation like I do.

I just want to create my art, not something else that comes pre-loaded on a tech company’s application. (That said, while they do have ready-made styles, the possibilities are endless when it comes to creating new styles and models.)

Now for a quick tour of some experiments I’ve done using my existing cartoons to teach the AI models, beginning with the most successful examples . . .

After “training the model” on 40 different background images taken from my animation, I told it to generate “A fiore cartoon of a partially destroyed city.”

Sadly, I made these before Israel started bombing Gaza — but used images from previous incursions into Gaza and other wars (including the Iraq War) to train the model.

To me, these look like something I would have drawn and the billowing smoke and telephone poles are almost exact replicas of my previous work.

My little AI bot seems to excel at creating images of destruction and oil refineries. (That probably says more about me and my cartoons than it does about artificial intelligence.)

I was able to generate a passable version of the US Capitol Building — but it does seem to have trouble generating things in my style that don’t have that bombed-out or industrial look.

I think the easier lift is teaching AI to generate static backgrounds for my animation, it’s much harder to train it to create characters, let alone actual animated characters that look like my work.

You can see some AI “animation” in this separate post . . .

See you back here soon, thanks so much for your support!

-Mark