A novel event reconstruction algorithm based on a Generative Neural Network is under development for water Cherenkov detectors, which have been one of the leading forces to understand neutrino physics and nucleon decay over the past decades, and will continue to do so in the foreseeable future. This novel technique shares a similar likelihood-based approach with the conventional algorithm currently in use in Super-Kamiokande (SK) and T2K, but with significantly fewer simplifying assumptions and more flexibility to address the vast complexity of such a detector. In addition to the remarkable reconstruction performance, the neural network in this work has shown a great potential of further improvement and broader applications. This poster presents on the construction, training, and performance check of several networks designed for the event reconstructions in SK.