Text this: Model-based portrait video compression with spatial constraint and adaptive pose processing.