After much messing around with several models lately. They can put anything together so long as you can find the basic components on google. Its like if you had a magic photoshop filter where you could copy two women and two men onto a desert background, throw some frog heads on the guys, stick a spaceship in there and have it all somehow blend together.

Where it falls apart is when you ask it to make something that can’t be assembled out of basic and well documented components. It can’t really do animal human hybrids unless they’re that kind or just a person with their limbs randomly turning into whole animals, it also struggles with changes to the fundemental structure of things so a humanoid alien with 4 arms is rarely going to work or even a vehicle with an unusual number of wheels for that vehicle.

Ultimately AI can remix existing art extremely well but it struggles to create anything it doesn’t already have hundreds of examples of.

Pretty awesome, actually kind of interesting the AI still hasn’t got depth of field down though. The buildings the hand is reaching for look like they’re really far behind it.