I find that i can spot AI Images fairly easily these days, especially the sort of fantastical tableaus that get posted to the various AI communities around lemmy. I’m tired of seeing them; it all looks the same to me. Was wondering if im being too sensitive, or if other people are similarly bored of the constant unimaginative AI spam…
For the record, I block any explicit AI Art communities that pop up in the feed, but there are more every day…
It doesn’t really bother me, but like you I am bored of it and I generally ignore it, or block communities if I’m seeing too much of it.
It is really cool that the models can generate fairly detailed images, but they’re all so similar and… boring. I once saw someone describe it like corporate art. It just tries to imitate something popular in a very mediocre way. You can keep re-training it, but it can still only imitate.
Still, if people are into it then that’s ok too. I have used it at work on occasion to create stupid little icons for internal tools I’ve built, so I guess there’s some little bit of utility.
My guess is that it’ll be used for a while for cheap and low effort branding, but soon companies will want to hire real artists again to differentiate themselves from the ML spam.
IMO, thats sort of the main use I see for AI image generation (and a lot of other “art”-AIs). There are plenty of cases where a graphic is needed that doesn’t need to be original, nor have any meaningful thought put into it. This could be a small icon that would normally be a free peice of stock art or programmer art, or it could be adding a unimportant backdrop to some character art that would otherwise just be left blank. Not all graphics have to be “art” and things that are “art” don’t have to be 100% original and hand crafted.
It’s because people are lazy. It needs extra work to generate something non generic. Also a lot of people using AI have no sense of beauty, as without AI, they are not very creative.
Using stable diffusion on a1111 myself, with controlnet, regional prompter, different checkpoints, a ton of Lora and inpainting, one can create much much better stuff. It’s not harder that way, just takes longer than copy pasting prompts and hitting the generate button.
I know this is true, because I see this daily by now. The amount of generic images uploaded to for example Civitai is proof of it.
It’s always interesting when they include specific details in the prompt that just get ignored!