cross-posted from: https://programming.dev/post/3974080
Hey everyone. I made a casual survey to see if people can tell the difference between human-made and AI generated art. Any responses would be appreciated, I’m curious to see how accurately people can tell the difference (especially those familiar with AI image generation)
The avocado had real text. Is Dall-E 3 capable of creating legible text?
Yes, it’s the only model that manages to get text right, and the results are usually pretty consistent. It’s a big step forward.
Base SDXL and SD1.5 with the help of controlnet can both do text too. I forgot Deep Floyd/IF can as well.
Control nets are kind of “cheating”, though, they’re a form of image-to-image where you provide them with something to trace over or otherwise guide them. I think in this area the open-source field has (briefly) fallen behind, we’ll need another round of catchup. That’s fine, though. Let competition drive hard.
It is, yeah
Kind of. It can generate readable text, but not all the time. It will frequently turn parts of your prompt into text that aren’t meant to be text or mix perfectly readable text with AI gibberish: