Hi, Francesco-
I think a good model/place to start for your example is with the "start-to-finish" example in the the AFNI Bootcamp. There is a two-task example there, for using multiple stimuli (2 there: an audio and visual). The afni_proc.py command specifically is (in the unzipped CD.tgz): AFNI_data6/FT_analysis/s05.ap.uber.
Therein, probably the "regress" block features there are the most useful, and you can make whatever contrasts are useful for your hypotheses. Some other blocks we would suggest other things, like how alignment to standard space is done (now often recommending an @SSwarper pre-step), for example. If you look at the @SSwarper help, you can see how that is integrated into the afni_proc.py command.
And feel free to ping back with any questions, too, of course.
--pt
