/ic/ - Artwork/Critique

>>6870309
it is still the same image. regardless of what file or compression you use, even if the underlying data differs.
this is NOT the case for latent diffusion models.
see pic related, see how all of these images look similar, but none of them are the same. they merely have the same style.

>>6870274
i can definitely see that happening since errors will pile up eventually. catastrophically so.
so i will probably concede that AI doesn't innovate beyond its boundaries.
but realistically, it still can innovate within its boundaries. and it does do that (by combining and mixing what it has, including styles).
and that also just means that original human input is still worth a ton, especially if you can give it your art as input.