Hi, Stef-
The motion censoring criterion is based on the enorm (Euclidean norm) of the motion parameter values; that is calculated using volume-to-volume changes (the derivative of each motion parameter time series), so that I don't think one fixed movement between scans should result in a huge amount of censoring (assuming that that displacement wouldn't lead to large outlier values for the "outlier frac" censoring criterion, which I doubt it would.
The afni_proc.py quality control (APQC) HTML output should show you where the censoring occurs in time-- you can see if it is where video clips are "glued together" or not.
If you look at the volumes in the AFNI GUI, can you see noticeable motion within a video clip's duration?
I don't think detrending separate from the rest of the regression model would be recommendable
--pt