At this point, I elect to ban all talk related to AI. I don't care if people like it or hate it. I don't care about the alleged metaphysical properties or lack thereof of AI generated works.
We occupy a very *weird* space, filled with weird people. People who have various values, outlooks, experiences, and understandings and for whatever reason, the mere existence of AI seems to amplify and agitate these various components.
I'm happy to let people enjoy things so long as what they enjoy isn't the cause of harm to anyone else. People should be allowed to be *weird*. But they also shouldn't have to view content they don't want to. Sequester the generated content as we do anything else and stop talking about it.
There's NOTHING productive about these conversations. There's no reason to have them and no value to be found.
skirtpie said: AI images have definitely surpassed real images.
I generate a lot of AI content and even I have to disagree. There are subtleties about a real human that AI isn't even close to copying yet. I will say that AI has definitely surpassed where it was a year ago. And just a few years ago we had to deal with six-fingered (or worse) people that AI would consistently generate. So, ya, it's come a long way.
As for 'surpassing' real images, I will say that I can create a wam scenario with AI that would likely surpass a human-produced one in the size of the mess, or the amount of messy destruction. Example: An AI high-diver does a huge dive and lands with a belly-flop into an Olympic-sized pool filled with chocolate pudding. Easy for AI. If a producer tried it, first of all, the diver would likely be injured by the impact, and possibly be stuck without air at the bottom. Next would the cost to fill that size of pool with pudding. Lastly, the liability, because this couldn't be done without a massive insurance coverage. So I will say AI surpasses human in this respect, the size and craziness of the wam situation portrayed. But for realism, I still prefer humans.
Interesting choice of subject line, given that "perspective" (or more accurately failure to properly understand perspective) is one of the way you can detect AI images from real ones.
In a real view of say a streetscape, all the lines will tend towards the same vanishing point in the distance. But to get that right requires a three dimensional understanding of space. AI generators often get this wrong, if you draw lines from things like path edges, buildings, etc, which should all meet at the vanishing point you find they miss each other.
Now TBH some models are better at it than others, and it's not unique to AI, here at the Hall we have an oil painting we bought back in the 90s of horses in a stable, it's a nice picture (which is why we bought it) but the parallax errors of the flooring in part of it are terrible - but forgivable in a £300 painting to support a local artist, and the horses and stable, the main parts, are gorgeous. But if someone presents a photo-realistic image, and the perspective doesn't quite work, chances are it's AI.
AI is fun and can produce some cool results if well engineered prompts are used, and can be an effective brain-trigger for arousal. But it doesn't yet come anywhere near what you get in a real image of a real person sploshing. One day it might, but it's not there yet, and if it does, it'll probably cost a lot more than it does at the moment. All that massive investment needs to be recouped somehow.
Love you, too
