Generating a series of images is an important task invarious fields of scientific research, e. g. Computational fluiddynamics (CFD). In the past years, solutions based on deepneural networks gained importance. In these tasks, it’s oftennecessary to declare regions of interest in the image. Further-more, the NN should only perform on these regions and therest should be ignored. With this paper, we propose an inno-vative and easy method for implementing this behavior in thefield of CFD.