Hi, Lena-
To add to Daniel's comment: would you be able to post your afni_proc.py command here? We might have some suggestions for that.
And also, as Daniel noted, for human data we lean on @SSwarper now pretty strongly for both skullstripping and nonlinear registration (in humans; for animal data, we would use @animal_warper---but I assume you are looking at human data at the moment). You would run that program before afni_proc.py, and then hand the results to afni_proc.py to use in processing. Is that what you are doing currently?
--pt