In these years, flow prediction using machine learning has been focused in the field of computational fluid mechanics. Because of the complexity of flow nature, eg. nonlinearity, divergency free, etc., successful cases are limited to the steady flow, or having to give some constraint of physical properties, known as PINNs. The present authors have been studying the prediction of incompressible flow using Convolutional Long-Short Time Memory (ConvLSTM) and Convolutional Neural Network (CNN) without any special constraint in order to apply various flow fields. The test flow is incompressible flow around a two-dimensional circular cylinder of high Reynolds number, shedding Karman Vortices repeatedly. The flow field are simulated with in-house software of incompressible flow solver using pseudocompressibility method and inner iteration preserving the temporal accuracy. Input data is one shot of flow properties of two velocity components and pressure for CNN and four sequential shots of flow properties, but the time interval is 100 times of the simulation time step. Input data area size is changed to examine the efficiency and the accuracy of the predicted results. The detail of comparison of predicted results will be presented at the conference.