look at the funky edges on the keyboard, the warped screens in the background and the cables and terminate randomly on the desk. this is an AI generated image.
But for me the most obvious tell is the edge of the desk. SD focuses too much on local details, so the foreground edge is not in-line with the background edge.
Anything with straight lines and precise angles is the nemesis of these particular algorithms.
Absolutely, yes. The screens generally look shopped in though but there’s that one angular AI-like glitch on the top one. The heap of backrests and screens in the back, right, also looks funky. Is that table floating? Also, that’s not actually the pope compare the ear.
I think it’s SDXL which actually got quite good at hands but that doesn’t mean that it suddenly knows what anything in any background looks like. Technical objects generally aren’t its strengths. The Eiffel Tower is always fun, it seems like it’s trying to make it into an organic canopy, if you want proper results for such things you have to train it first (or download a lora).
look at the funky edges on the keyboard, the warped screens in the background and the cables and terminate randomly on the desk. this is an AI generated image.
The more you look at it the worse it gets.
But for me the most obvious tell is the edge of the desk. SD focuses too much on local details, so the foreground edge is not in-line with the background edge.
Anything with straight lines and precise angles is the nemesis of these particular algorithms.
Absolutely, yes. The screens generally look shopped in though but there’s that one angular AI-like glitch on the top one. The heap of backrests and screens in the back, right, also looks funky. Is that table floating? Also, that’s not actually the pope compare the ear.
I think it’s SDXL which actually got quite good at hands but that doesn’t mean that it suddenly knows what anything in any background looks like. Technical objects generally aren’t its strengths. The Eiffel Tower is always fun, it seems like it’s trying to make it into an organic canopy, if you want proper results for such things you have to train it first (or download a lora).