No, I do not generally agree that AI-generated images are “theft”. GiovanH’s blog post explains it better than I could - please go read it when you find the time. But a tl;dr is that models aren’t simple collage machines - they actually pick up concepts from the images they’re trained on, and demonstrate an ability to combine them to output novel ones - not exactly, but pretty similar to how concepts are combined in manually-made art. It’s also mathematically impossible for image diffusion models to directly contain their training imagery, due to their small size (SDXL models, for instance, are around 6.5 GB while being trained on billions of images). Of course, there is a small chance of overfitting happening, which the post gets into more detail about.
Also, I don’t believe it’s meaningful to distinguish whether something has “soul”, based on its medium. People can’t agree on what a “soul” is, or if it even exists. What can actually be quantified is whether AI art invokes reactions in people - which it most definitely does, whether you find the comics funny, are repulsed by their mediums, or simply shrug at them and move on. Besides, it was a human who prompted to generate the image in the first place.
AI is a new technology, and it’s totally OK to be worried about its impact on society. However, I’d say the best way to go about it is to clearly state each other’s opinions and skip the buzzwords and assumptions. If you’re willing to reply back, feel free, even if you find yourself disagreeing most of the time! As long as we can keep this civil.
It steals the art of real artists to generate its images.
So yeah it steals and projects their soul too.
It’s gross and its creator should have to pay every artist it stole content from to train their bot.
Piracy isn’t stealing.
Stealing from artists and not corporations is stealing.
Also trying to pass off other people’s art as your own is not piracy.
Alright, I’ll make my point seriously.
No, I do not generally agree that AI-generated images are “theft”. GiovanH’s blog post explains it better than I could - please go read it when you find the time. But a tl;dr is that models aren’t simple collage machines - they actually pick up concepts from the images they’re trained on, and demonstrate an ability to combine them to output novel ones - not exactly, but pretty similar to how concepts are combined in manually-made art. It’s also mathematically impossible for image diffusion models to directly contain their training imagery, due to their small size (SDXL models, for instance, are around 6.5 GB while being trained on billions of images). Of course, there is a small chance of overfitting happening, which the post gets into more detail about.
Also, I don’t believe it’s meaningful to distinguish whether something has “soul”, based on its medium. People can’t agree on what a “soul” is, or if it even exists. What can actually be quantified is whether AI art invokes reactions in people - which it most definitely does, whether you find the comics funny, are repulsed by their mediums, or simply shrug at them and move on. Besides, it was a human who prompted to generate the image in the first place.
AI is a new technology, and it’s totally OK to be worried about its impact on society. However, I’d say the best way to go about it is to clearly state each other’s opinions and skip the buzzwords and assumptions. If you’re willing to reply back, feel free, even if you find yourself disagreeing most of the time! As long as we can keep this civil.