Hello, astrophotographer here.

The unique thing about astrophotography is that the subject that we are imaging is (relatively speaking) static. Any slight deviation, even the smallest detail, that is not structurally accurate is PAINFULLY obvious.

As an aside, we do use traditional neural networks for certain processing steps. Deconvolution and noise reduction are very common. However, these are only helping in correcting errors caused by viewing conditions. They are not the "imaginative" AI that you are dreading.