Under the current interpretation of the law, you have to make modifications to the ai generated image in order to retain some form of copyright over it.

This would be an ideal test case for the law. If an artist uses a computer as a tool in the creation of a work, then they retain copyright. Consider a 3D animation. The artist does not manually and with intent create the highlights, shadows, and (often) textures, etc. They are emergent properties of the way the computer generates the image based on the artist’s inputs.

Lots of fine artists use computer generated elements, such as 3d printed patterns, photo collages, iterative designs, etc.

It seems that the legal question would be whether generative images are categorically distinct from other types of computer aided creation. It’s hard to imagine an argument that an artist created the data, wrote a program to use the data to train a model, used the model to generate an image, created some type of prompt, and published the output… but doesn’t own the output. It would be an interesting argument anyway.

Are you talking about an artist exclusively running their images through an AI model until it is capable of regenerating images that look like it was created by them and have some semblance of intent?

In order to get anything that looks remotely like what people want, I’m pretty sure they would have to upload millions of pictures of their own creation first. So most people just layer their images on top of the giant mash of ethically sketchy data that already was there.