I have a video going over it. But basically the idea is you train the algs on a virtual cube so you memorize it that way and can see how the pieces are moving and stuff. Then you can do it on an actual cube. I find it transfers pretty well since you're going to be slow at executing new algs no matter how you learned them. This just makes the learning process faster.
No you should first learn the keyboard controls until you aren’t thinking about them. Then you learn the algs by reading it and executing it (which you shouldn’t have to think about) so you can watch how the pieces are moving on the cube and learn it that way. Then when you do it in a real cube you will be watching the same patterns.