We propose a causal generative model that accurately estimates a broad class of causal queries, including counterfactuals, in the presence of hidden confounders.