But again, I think the idea that the model weights store the source images as if it were a library is fallacious. There. Is no known way to get it to spit back

But again, I think the idea that the model weights store the source images as if it were a library is fallacious. There. Is no known way to get it to spit back out a copy of a training image.