Copyright law generally protects the fixation of an idea in a “tangible medium of expression,” not the idea itself, or any processes or principles associated.

Neural networks don't store images inside, they decompose these images into elementary concepts and then recompose new images from such concepts. Basically they learn the unprotected part of the training set.

Think about it in size: 4 billion images shrunk into 4GB, that means a measly byte per input image. Not even a full pixel! It certainly has no space to store those images. It can only store general principles.

Getting offended for having a single byte learned from one of your images seems unjustified. On the other hand it looks ugly how pre-AI artists are gatekeeping the new wave of AI assisted artists. Let people eat cake.