We design nearly optimal differentially private algorithms for learning two fundamental families of high-dimensional distributions in total variation distance: multivariate Gaussians in $\mathbb{R}^{d}$ and product distributions on the hypercube. The sample complexity of both our algor