The approach does make sense to some degree. It's usually better/easier to align the epi data to the anat of the same subject and then apply the additional transformation to a standard space computed from @auto_tlrc or auto_warp.py/3dQwarp. See the "Bells and whistles" example in align_epi_anat.py's help. The mask data should be transformed using a nearest neighbor interpolation instead using "-final NN" .

The error you got is a bug in applying the AddEdge option to NIFTI datasets. I'm working on fixing that, but in the meantime, use an AFNI format copy as the input data to align_epi_anat.py. Thanks for letting me know.