Modern numerical weather prediction models are amazing: amazingly powerful, amazingly accurate – amazingly complicated, amazingly expensive to run and to develop, amazingly difficult to use and to experiment with. We use them in historical climate reconstruction, to make reanalyses, but I’d rather have something less amazing, but much faster and easier to use. Modern machine learning methods offer sophisticated statistical approximators even to very complex systems like the weather, and we now have hundreds of years of observations to train them on. This means we can use a Variational AutoEncoder to build a fast deep generative model linking physically-plausible weather fields to a complete, continuous, low-dimensional latent space. This model is fast and easy to use, but powerful enough to be used for data assimilation or as a GCM.