> Erik Wallace, a computer scientist at the University of California, Berkeley, told me in a phone interview that the lawsuit had “technical inaccuracies” and was “stretching the truth a lot.” Wallace pointed out that Stable Diffusion is only a few gigabytes in size—far too small to contain compressed copies of all or even very many of its training images.
>In reality, Stable Diffusion works by first converting a user’s prompt into a latent representation: a list of numbers summarizing the contents of the image. Just as you can identify a point on the Earth’s surface based on its latitude and longitude, Stable Diffusion characterizes images based on their “coordinates” in the “picture space.” It then converts this latent representation into an image.
Regardless of the lawsuit, this seems like a quite spurious, self-serving argument in and of itself. If the “coordinates” of the image were not equivalent to the image, how would it be possible to fine tune the model on a set of images and recreate their style? This is setting aside the myriad examples where SD regurgitates fairly close copies of individual works.
This is a bit like training a model on the works of Shakespeare, but then claiming that the tensors resulting from ingesting the data are not actually Shakespeare’s writing. Oh really, where’d they come from then? Thin air?
>In reality, Stable Diffusion works by first converting a user’s prompt into a latent representation: a list of numbers summarizing the contents of the image. Just as you can identify a point on the Earth’s surface based on its latitude and longitude, Stable Diffusion characterizes images based on their “coordinates” in the “picture space.” It then converts this latent representation into an image.
Regardless of the lawsuit, this seems like a quite spurious, self-serving argument in and of itself. If the “coordinates” of the image were not equivalent to the image, how would it be possible to fine tune the model on a set of images and recreate their style? This is setting aside the myriad examples where SD regurgitates fairly close copies of individual works.
This is a bit like training a model on the works of Shakespeare, but then claiming that the tensors resulting from ingesting the data are not actually Shakespeare’s writing. Oh really, where’d they come from then? Thin air?