What is your AFNI version (afni -ver)?
For your use of 3dLMEr, I recommend the following model specification:
-model 'condition*timepoint + age + gender + movie + (1|Subj) + (1|Subj:condition) + (1|Subj:timepoint)'
For more details on specifying individual-level random effects in hierarchical models, see this AFNI blog post.
Regarding effect estimation: for a contrast like this,
-gltCode swear_words_time0to2 'condition : 1*swear_words timepoint : 0.333*0 0.333*1 0.333*2'
you might consider using -glfCode instead, which may offer greater sensitivity. For example:
-glfCode swear_words_time0to2 'condition : 1*swear_words timepoint : 1*0 & condition : 1*swear_words timepoint : 1*1 & condition : 1*swear_words timepoint : 1*2'
You can apply a similar strategy to your other -gltCode specifications as well.
Gang Chen