We develop a method that can be used to turn any multi-layer perceptron or convolutional network into a normalizing flow. In some cases this requires the addition of uncorrelated noise to the model but in the simplest case no additional parameters. The techniques we develop can be applied to a broad range of transformations. Converting standard models to normalizing flows allows the same architectures to be used for a wide range of tasks. Our models also allow existing density estimation techniq…