I'm assuming that by "I had to repeat each covariate 4 times to match each emotion condition within the same subject" you mean that the covariates vary across subjects but not across conditions.
A comment: it looks like you don't have subject effects modeled. The good thing is that if did model them, it would be tricky to assess the effects of the covariates (because they're constant on a given subject), though maybe you could do so using the ideas of the Glascher/Gitelman monograph people refer to on this list. The bad thing is that you lose power by not modeling subject effects. Though if the scans you brought to the second (group) level are already differences at the first (subject) level, then one could argue the subject effects have already been subtracted out, I believe.
It seems like design 1 models only the main effect of each covariate, whereas design 2 models the condition X covariate interaction. If you really want to look at the interaction, and believe it's not insignificant, then you should use model 2. The one disadvantage of model 2 is that it's difficult to consider the main effect of condition alone.