Normal facesets are set up as 4x2. So, 4 actors in the top row, 4 actors in the bottom row, and each face is 96x96. So, when you are normally selecting a face, you are selecting one of those eight faces. That is what I was referring to by select the index.
For large faces, you can do it in one of two ways, either you can make a 4x2 set of faces that are larger than normal (ie. 4 faces in top row and 4 in bottom, but instead of being 96x96 they are 160x128, or whatever). Then you select the face you want, but if I recall correctly, the selecting square will not be aligned properly, so that's why I said index instead of face.
The other way to do it is to just save single large faces with the $. What that $ means is that the script will read the image as just one face, rather than 8 faces like normal.
If you are still confused, a large face is used in the demo. You can just look into that event and see how it was done.