Is it fails, if asked for "one on top" as well?
If yes, then can you also try "Domino with 2 spots and 1 spot" or "Domino 2 and 1"?
Interesting. It's actually much worse than I expected it to be. Maybe there was some sort of cleaning to remove duplicate images from the dataset.
A few more requests, I would really like to see if you decide to do them.
"Simple red dice showing six on top"
This is to see whether other dice sides would be coherent with what's on top.
"Very cool car"
This one is tongue in cheek to see whether it would generate a frozen supercar to maximize both meanings of "cool".
I would be interested in two kinds of prompts:
First, can it reproduce something really popular like:
"V-J Day in Times Square - Alfred Eisenstaedt, 1945"
I know, that original has some faces, so it would be impossible to share, but still interesting to know the result.
Second, does it know some of the not so mainstream video game "styles"? Screenshots from any of the following would be perfect: "Don't starve", "Heroes of Might & Magic III", "Sid Meier's Civilization III", and "StarCraft".
Huh, it really can't do the math. I wonder if Flamingo is any better at it.