Starnet++ is an application that uses machine learning to remove stars from astronomical images, mostly without leaving unsightly artefacts.

In processing astroimages it can often be a challenge to edit in a way that gives good results with both stars and more diffuse objects like nebulae or galaxies. One way to deal with this is to make different edits for each, and then recombine the images to achieve the best of both. One of the steps in doing this is to remove the stars from the image to enable the rest of the image to be adjusted to show as much detail as possible. With the stars left in they often become bloated, and even more often end up being pure white due to the brightness being increased to a point where no differentiation of the colour channels remains.

After processing the starless image separately from the one with stars included, the two can be recombined in a number of different ways in an attempt to produce a pleasing combination. I still consider myself to be very much in a learning phase in image processing, but I think that my most recent images demonstrate that I'm getting a better grasp on how to do this reasonably well.

There isn't a lot of information on the Starnet website, but it's worth a visit anyway.