Kallai,
First of all, even though very popular, "regressing out" a variable is not an accurate characterization when you account for the effect of the variable. Instead "controlling the effect of the variable" at a specific value (usually, but not necessarily always, the mean) is a better description.
> The first sub-brick [#0] seems to be identical to an analysis of stim_times without the AM2.
> Does it mean that the second sub-brick [#1] is the effect after regressing the RT out? If not
> - is there another way I can control for different RTs?
The first sub-brick [#0] is response magnitude of the condition controlled at the average of the modulation variable (RT in your case), which is why you see similar results to the original analysis without modulation. The second sub-brick [#1] is the modulation effect of the variable. That is, it shows the amount of response when the modulation variable increases by one unit (e.g., when RT increases by 1 msec).
Gang