Each image in the training set contributes only about a single byte to the AI model. Meaning there are no “source” images in any meaningful way.
Edit: A minute of fiddling with DALLE-3 I got this (and a dozen similar ones). Did I stole from OP? Did we steal from the same source? Or is a prompt like:
“cool sky wizard with a long lush white beard with sunglasses and a Gandalf hat shooting an ak-47 in front of a galaxy cloud scifi backdrop, fog, clouds, magical, stylized digital painting, heroic posing, god rays, forshortening, roto zoom, wind blowing through his hair, brass flying out of the gun”
Simply enough to describe this kind of image reasonably closely?
Doesn’t change the fact that these apps are trained on stolen data, and soon, when they’re sued into oblivion, they’ll be forced to credit artists in a meaningful way.
When artists can’t even figure out what was stolen they really have no leg to stand on. Complete cluelessness of how any of this works doesn’t help either.
And beside these kinds of arguments are just complete hypocrite bullshit to begin with. Human artists use references all the time. If the law decides that that’s illegal, they’ll be in even deeper trouble than they already are.
Also the only one that would profit from these lawsuits would be Adobe, Shutterstock and all the other mega-corp content hoarders out there. You’ll be stuck with AI image generation all the same, but instead of it being free and available to everybody, you’d be paying a hefty subscription fee to them.
Either way, AI image generation is the way forward. The genie is out of the bottle. Get used to it
Each image in the training set contributes only about a single byte to the AI model. Meaning there are no “source” images in any meaningful way.
Edit: A minute of fiddling with DALLE-3 I got this (and a dozen similar ones). Did I stole from OP? Did we steal from the same source? Or is a prompt like:
“cool sky wizard with a long lush white beard with sunglasses and a Gandalf hat shooting an ak-47 in front of a galaxy cloud scifi backdrop, fog, clouds, magical, stylized digital painting, heroic posing, god rays, forshortening, roto zoom, wind blowing through his hair, brass flying out of the gun”
Simply enough to describe this kind of image reasonably closely?
Sure Jan.
Doesn’t change the fact that these apps are trained on stolen data, and soon, when they’re sued into oblivion, they’ll be forced to credit artists in a meaningful way.
When artists can’t even figure out what was stolen they really have no leg to stand on. Complete cluelessness of how any of this works doesn’t help either.
And beside these kinds of arguments are just complete hypocrite bullshit to begin with. Human artists use references all the time. If the law decides that that’s illegal, they’ll be in even deeper trouble than they already are.
Also the only one that would profit from these lawsuits would be Adobe, Shutterstock and all the other mega-corp content hoarders out there. You’ll be stuck with AI image generation all the same, but instead of it being free and available to everybody, you’d be paying a hefty subscription fee to them.
Either way, AI image generation is the way forward. The genie is out of the bottle. Get used to it