Lets say our frame 0 pic is 40x40 pixel, pics for eyes and mouth both 20x10 each.
Eye pic is at correct position at 10 px distance from upper(=Y) and left border(=X) of frame 0 pic,
while mouth is placed correct at 25 from upper and 10 from left.
If we have correct coordinates for this or any other relation of frame0 pic to rest in .sti,
we can calculate offset values for any other dimensions of frame0 pic
or the dimensions of other pics in same .sti file for face anim.
So its probably best to keep the upper left corner of frame1+ pics at the same spot in all portrait pics of same dimension,
and use equal eye%mouth values in .xml for at least the new entries
either use always the same.

There are a couple of tools in the Gimp(and probably most other pic editors)
which allow to do at least the general work faster like blur&unsharp mask, random pixel distortions ...

OT question: Anybody knows where the 106x122 "photos" in this thread were taken?