It is a neural-net program, written in Python, that processes 8-bit TIFFs to remove stars. It's pretty wild. I had a little trouble getting the prerequisites working on my Mac but basically you
go here and follow the instructions. The trainings are a separate download, if you neglect those you'll get an obscure message about how it's unable to load its previous checkpoint or something like that.
It yields a version of your image with the nebulosity intact, but the stars missing. As Pete notes, you can then apply pretty vigorous modifications without worrying about bloated or blown-out stars.
If open the starless image in Photoshop as a layer above the original, you can temporarily set its blend mode (right-click on the layer) to "difference" and the stars will display all by themselves. Then you can use Photoshop's "stamp visible" command (on my Mac it's shift-option-command-E) to create a new layer with just the stars in it. Tuck that away at the bottom of the stack, set the starless layer back to "normal" blend mode, and go to town on the nebulosity. When you're done with that, do the same Stamp Visible trick to the enhanced starless image so you have an easily-portable capture of your manipulations. Put the star layer on top, perform any manipulations such as saturation enhancement or sharpening, Stamp Visible one more time. Put the starless Stamped Visible at the top of the stack, put the stars above that, set the blend mode on the stars to "Lighten". Pete's probably got the better technique for controller the starriness in the result; you could play with the star layer's opacity, or edit it with the Dust and Scratches filter if you prefer.