You made me curious so I googled. The results were shit. My wild hunch - based on no facts - is that AI creates images by sort of "layering" stuff and hoping for the best. In the image there's...

The court
The guys in the background
Trump & Hochul & the ball
The net

Everything is best guess. In other words AI uses the following "logic" and works from there.

Basketball courts are ~100 feet long.
People in the background will be smaller than in the foreground.
People are - I guess - about 5' 5" on average? Male basketball players are much taller of course.
The net is bigger than the ball.

Whatever is going on - there's a sort of - I don't know how to describe it - weird smoothing that beyond the mistakes creates a sort of generic perspective. I have a hunch that if you used a similar prompt but added - give it a raccoon's eye perspective from on the court - the results would be garbage because on the net there are no images of people playing basketball in a stadium from that POV and the AI wouldn't know what to do.

Rant. Can't there be a free search engine that's not garbage?

[–] gasgiant@lemmy.ml 7 points 3 days ago

There's another alternative that gives AI even less credit and how I think it probably really works.

The image is just an average of all the slam dunk on player images it has access to with faces and some details pasted over it.

Your method requires a lot more calculation based on understanding the composition of the photo, depths, courts, players etc...

Whereas making an average of all the shots it has learnt that are like that requires much less work than trying to design it completely by knowing about all the composite parts.

Which is what gives all the smoothing and not quite right feel for it.