The results so far look promising! I'm curious if this technique can be generally applied to any arbitrary compositing method (like something more ad-hoc than s

The results so far look promising! I'm curious if this technique can be generally applied to any arbitrary compositing method (like something more ad-hoc than screen and multiply) given the proper training set.