Agreed, you can tackle one problem at a time. And I think progress on one might spread over to the other also.

A k2+ gender version would be an interesting expansion on the idea though. Since the notion of a complementary or inverse gender isn't universal to all genders, maybe cases where there's no clear inverse the gender could be swapped with a randomly chosen gender. Or just swap any explicit gender reference with a different randomly chosen gender. This becomes a bit more challenging if you want to keep the output cohesive though, since you'd probably have to introduce some sort of entity linking scheme.

Another issue is finding an appropriate k for genders, assuming this package is implemented with some sort of hard coded gender mapping. This becomes more complex when accounting for different cultures, which probably just ends up multiplying k by m cultures.