To more accurately capture differences in the hemodynamic response between the two conditions, I recommend using the following model specification with 3dLMEr instead of 3dLME:
3dLMEr \
...
-model "condition*timepoint + age + gender + movie" \
...
In the output, the interaction term condition:timepoint primarily reflects the differential hemodynamic response between the two conditions.
Alternatively, you may consider using 3dMSS, as described in this AFNI blog post and the associated paper, which focuses on enhancing sensitivity to HDR differences through profile estimation.
Gang Chen