If you train it on graphics from magic cards I'm sure it will generate things that look like graphics from magic cards. The challenge is to get it to create some kind of new artistic value. That parkinson one of the EQ chick literally just looks like it scanned and translated the pixel order of the actual image, more or less. That's kinda like ehh ok.
It's fun to reverse engineer it conceptually. Like there are lots of apps that let you generate music in various styles, and it's really fun and easy to be like "make a song about eating grapes in the style of the beatles" and it will spit something out that sounds more or less like what you asked for. But then ask yourself what prompt would be required to get it to generate an actual beatles song that already exists? Like pretend the actual human beatles never existed, and your task is to get an AI to generate the specific song penny lane, what would the prompt be?
TLDR: This stuff is remarkable in many ways, but not actually intelligent in any meaningful sense.
|