If the marginals have the normal (gaussian) density wrt Lebesgue's measure, it's easy since we know the pdf of a bivariate gaussian distribution

I don't think it's easy to characterize, there is already such a wide range of possible marginal measures...