I have seen a lot of articles and social media pushes to stop AI image creation because it can be trained to the style of a specific artist. It's constantly about how the poor artist won't be able to make mo ey because the AI can do their art for them.
I doubt this. Artists have multiple styles and are more well known for the story within their pictures. If I hired an artist who could repeat that style and do something similar then it's ok?
This makes no sense. Instead, I bet it's a big business trying to protect itself. Disney has a full department that decides on styles for art and presentations. Genie must look this way in all pictures and all artists must repeat it perfectly. Only Disney can sell products with this genie or anything close to it.
If I had AI make Genie doing something and then printed that out, there is very little Disney could do to stop it. This is the music industry vs Napster all over again.
Usually legally distinct, but not always. I've seen examples where image AIs have included watermarks from whole sections of source image and code bots that include functions verbatim from copyrighted code.
The larger the model the better the results are, but also the more is memorized instead of interpolated.
When they get up to trillion parameter models you'll be able to describe some obscure picture you saw on the internet and it'll recreate it perfectly.
I just saw a funny example of that the other day.
The person who posted it pointed out it's "The one time it creates legible text."
Here's another one. Still can't do hands but we've got watermarks down to a science.
I'm morbidly curious what will happen when people start injecting enough AI generated images into the sample pool that you start seeing the AI version of inbred genetic defects.