> Is it the correlation between Face+base+orig and seed.1D?

Yes, just like [afni.nimh.nih.gov]

> And you say "maybe the step of regressing out those unrelated regressors is
> not good enough because that assumes a perfect response at EVERY event.",
> which means the 3dSynthesize can't really exclude those unrelated conditions,
> right?

Yes, that's my guess.

> The order of polynomials for baseline drift. I follow the defult 3, but am not
> sure what number I shold use.

The order of polynomials for the drifting effect decides the cutoff frequency at the lower end. A rough rule of thumb is one order every 150 seconds of data.

> Number of parallel jobs. Actually I completely do not know what is that mean....

This is for parallel computing. If you have multiple processors on a computer, you can take advantage of that by specifying the number of processors.

Gang