Re:Stable Diffusion Frivolous Litigation
by visarga on Sunday January 15, 2023 @02:00PM
(#34392007)
> Nobody cares about the model output
> The issue at hand is that artists had their art trained on without their permission
Interesting that you care about the training data but not the generated images. The only explanation I found is that you want to stop the "evil" from the root and have the models unaware of your style.
Do you believe that the style of an artist who is not included in the diffusion training set can't be replicated? No style is actually outside the scope of the model after it has learned the first few thousand styles. They become a map that will reveal every other style.
> The issue at hand is that artists had their art trained on without their permission
Interesting that you care about the training data but not the generated images. The only explanation I found is that you want to stop the "evil" from the root and have the models unaware of your style.
Do you believe that the style of an artist who is not included in the diffusion training set can't be replicated? No style is actually outside the scope of the model after it has learned the first few thousand styles. They become a map that will reveal every other style.