With ChatGPT’s new image generation capabilities, I decided to play. Specifically, I wanted to see how weill it would take this image:
from the Ask Leo! home page, and re-render it in different styles. (Click on any for larger, if you dare.)
“Make a 3-d animation style image of this headshot”
I can see myself in it, but it definitely feels not-me as well. What I find interesting is the level of detail recovered in the background. You can’t tell from the original, but that is indeed Windows XP’s Bliss in the background.
“Now make a wall street journal line art style version of the original image”
A good capture of WSJ style art, for sure. Still both me and not-me.
“Now make a sunday newspaper comics art style version of the original image”
OK, that’s just … wrong. The art is peachy, but how I got the crazed look based on the starting image is disturbing.
Time to change tactics and image.
Another headshot
A more traditional, if boring, headshot.
Better, but I still find the subtle changes a bit much. Once again, it’s me but it’s not me. Perhaps more me than the previous images, but nonetheless, still … very-close-but-not-me.
So I decided to lean into the not-me-ness…
“Create a street art caricature of this image”
Not me. Interesting but not me.