Traditional water body generation methods often need to study the law of wave motion, and can not generate corresponding three-dimensional water body according to two-dimensional photos. In order to solve the above problems, a 3D water generation method based on 3D Gaussian is proposed. FLIP was used to calculate the force on the fluid with the grid, interpolate the acceleration onto the particles, and make the particles move according to the speed. The physical parameters of water movement were integrated into the static water scene generated by 3DGaussian to realize the dynamic water generation. This method demonstrates the ability to create diverse water content, and has certain reference value for computer water generation methods.

dynamic scene reconstruction, water generation, deep learning, FLIP