If there are several runs in an fMRI dataset, AFNI seems to add a different set of detrending regressors for each run, but there is only one set of activity regressors and one set of motion regressors. SPM, on the other hand, seems to add one set of activity regressors and one set of motion regressors per run.
I'm not saying that either way is wrong, but it would be interesting to know why AFNI uses the first option.