This repository contains an Auto-encoder ConvLSTM network (Pytorch) which can be used to predict a large number of time steps (100+). The network prediction is sequence-to-sequence which works well to predict 5 to 10-time steps in one pass through the neural network. The network is trained for unsteady fluid simulations using data. Another training method tested is the physics constraint method, where governing equations of fluid motion are used to optimize loss. Few attempts to train unsteady Navier-Stokes are made, but it dint work.