AI Art Soup

There is such a thing as a World Record for the "Fastest Drawing of Fred Flintstone". To date, artist Lev Cantoral still holds the record at drawing a recognizable Fred Flintstone in 6.27 seconds.

I didn't exactly time my effort at using AI to create a drawing of Fred, but I would say it took half a minute to type in a prompt and have the software "generate" an image:

Lev has nothing to worry about.

  This was more Ood Abbad Abbay than Yabba Dabba Doo. There is a feature where you can upload an image of Fred Flintstone and the AI remakes it into whatever it thinks it is, but that's not quite as fun...plus, that's the element that makes established artists nervous - the chance of someone selling reproductions of someone else's art, so I stayed away from that. I had more fun seeing what the AI would generate from scratch anyway.

That was supposed to be "Scrooge McDuck drawn by Chuck Jones"...I don't think the software has a grasp of Chuck Jones' style much.

This "technique" is called Stable Diffusion", or "Text-to-Art", the flip side of using talk-to-text. This actually combines capabilities that have been in existence for over a decade: photo filtering, image search, text search prompts, etc.. the results are a lottery...

"Scrooge McDuck in Disneyland"

"Ducktales"

"Kevin Conroy as Batman"

"Full-Body of Batman, drawn by _____."

            Pokies, huh? AI did it.

"Full-Body of David Tennant as Constable Hamish Macbeth, in uniform, overlooking a field in Scotland."

    AI don't know what a Scotland 

    Constable's uniform looks like.

"Lindsay Lohan as Batgirl, as drawn by ______"

                            Ok...

"Leslie Grace as Batgirl, as drawn by ______"

                       Hmm...

"Karen McDougal as Batgirl, as drawn by _____"

        That's...really, really good!

"Lily Collins as Batgirl, as drawn by _____."

    The likeness is very on-point.

"Barbara Gordon as Batgirl, as drawn by _____."

I found it interesting that instead of a mask, it generates glasses or bifocals, while the hair and cowl  form one piece or "helmet" shape, while the facial features are approximate to mimicking the artist I had in mind, but not quite; it might as well be the AI as "student" learning to draw from an influence. Meanwhile, the bat-emblem is a puzzle for it; the one with McDougal created a Power Girl-esque "Boob Window" without revealing anything. The ones with Lily Collins are clearly dipping into photo references, but a lot of flesh-and-blood comic artists often employ the same trick. Also, there's never a repeat of the same costume in each image, although when I specifically wanted Lily to wear a purple costume, then the color was consistently the same, as well as the color of her hair. The eye color was variable.

At this point, I exhausted the free trial on the site I used and wasn't interested in paying to continue, but I thought it was fun - not quite 'there' in meeting up with human imagination and creating images we're thinking, but the limitations are rooted in our ability to articulate what we want exactly. Once the developers crack that, the sky's the limit.