The paper introduces Real NVP (Real-valued Non-Volume Preserving transformations), a set of invertible and learnable transformations designed to extend the space of probabilistic models. These transformations enable exact log-likelihood computation, sampling, inference of latent variables, and an interpretable latent space. The authors demonstrate the effectiveness of Real NVP in modeling natural images on four datasets: CIFAR-10, Imagenet, LSUN, and CelebA. The model's performance is competitive with other generative methods in terms of sample quality and log-likelihood. Real NVP bridges the gap between auto-regressive models, variational autoencoders, and generative adversarial networks, offering a flexible and tractable approach to unsupervised learning. The paper also discusses the use of batch normalization and other techniques to improve training and performance.The paper introduces Real NVP (Real-valued Non-Volume Preserving transformations), a set of invertible and learnable transformations designed to extend the space of probabilistic models. These transformations enable exact log-likelihood computation, sampling, inference of latent variables, and an interpretable latent space. The authors demonstrate the effectiveness of Real NVP in modeling natural images on four datasets: CIFAR-10, Imagenet, LSUN, and CelebA. The model's performance is competitive with other generative methods in terms of sample quality and log-likelihood. Real NVP bridges the gap between auto-regressive models, variational autoencoders, and generative adversarial networks, offering a flexible and tractable approach to unsupervised learning. The paper also discusses the use of batch normalization and other techniques to improve training and performance.