The photography example is a good one imo. There's a big difference between pointing your camera at something and snapping a quick photo, vs carefully setting a scene, lighting, and adjusting a lot of camera settings before finishing in Photoshop.

In comparison, you have an image generation where someone types cat into midjourney, vs someone in stable diffusion blending 3 models, using latent coupling to compose the layout of the whole image, using control nets to position all the individual people, and finally touching the image up in Photoshop.

I have no doubt that higher effort photography or high effort ai art should be covered by copyright, each can take hours of work to produce the desired result. It's just not feasible to verify how high effort the production of something is for general copyright.