You're viewing a single thread.
Someone explain to me why i have a sense for ai? This is like the fourth time this week i see an ai image and i spotted it even tho at first i hadnt seen the mistakes the ai made. Something just feels off in ai pictures and i cant really pin down what it is. But when i see one i can just spot it for some reason.
There's also some artifacting around the lines that are reminiscent of JPEG, but not kind of that, and it seems like the text is superimposed onto it in a second or third pass by an external tool to make it look more consistent.
Uncanny valley. The AI gets things right overall, but not enough that you don't realize somerhing is wrong.
For me it's the arms being fused with the table,pessimist only having one arm, the faces that feel too generic/soulless (like, they barely change between pesimist and excel), the "optimist" being the glass and not another character...
The glass is also weirdly large considering it looks like it should be at roughly the same distance as the people.
The glass is also speaking, as opposed to a third person.
I noticed that the new image model doesn't like pure white backgrounds.
It's become such a reliable tell that I'm genuinely curious why AI seems to think that cartoon images MUST look like they're drawn on old paper.
EDIT: The AI answer I got when asking Google was about what I'd expect, but this last bit I hadn't considered:
Sepia tones can help unify the color palette of an image, making it more visually cohesive and pleasing. This could be a factor in the AI's decision to use sepia, as it can contribute to a sense of order and harmony in the generated image.
So it uses it to try and make it more visually appealing to a wide audience, leading to it looking even more bland and uninspired.
How do you determine that something is AI?
The uncanny valley. It looks overall right, but some details don't feel quite right even if you dln't know which ones.
So, vibes basically.
Yeah, until you find what's actually wrong, like the arms that fuse with the table, the talking glass, the almost identical faces, the glass being giantic, the backgroud colour, the specific artstyle (just like when AI made images that looked too shiny and shit, and it was obvious it was AI)...