pull down to refresh

I would even say those that are shocked by DALL-E haven't spent a lot of time trying to get it to do a very specific thing.
Like, try to imagine a fictional character or creature and make it as detailed and specific as possible (pose, clothes, accessories, features, viewer perspective, etc), and try to get AI to generate it exactly how you envision it, it turns out to be way more difficult than you might think. Maybe I'll post a bounty with some specifications for an image idea just to try to prove that point.
50 sats \ 1 reply \ @Atreus 5 Mar
I gave up on it. It's good at getting “an image,” but not the image I have in my head. Unfortunately, it's become harder to source man-made images since DeviantArt, ArtStation, et al., are overrun by AI generations.
I've been looking for a good place to source human-made scifi/cyberpunk images for my newsletter and it's annoyingly difficult.
reply
It's good at getting “an image,” but not the image I have in my head.
exactly
reply
Here ya go: #451746
reply
Fully agree. I was confronted with its limitations when I was following my son's instructions to make the robot he had in mind. As he does not understand the underlying tech, he was the most critical customer one could imagine.
E.g. he was asking the robot to have square eyes (as that's what he saw in a robot from the toystore), yet the generated images kept showing round eyes. Even though ChatGPT explicitly told me "I took special care of including square eyes as you asked". And if I ever managed to include something my son asked to change, it changed many other things as well. Understandably, as that's how it works, but a hard sell at this point to convince this tech can fully replace humans.
My son will not be returning customer for now :)
reply