The manuscr ipt describes the design of a new codec WZVC, where SI generation performed on unsupervised learning of two motion fields.