This project explores hybrid-neural representations for complex editable object-compositional urban environments. It expands controllability in the latent space by enabling global urban-level interpolations, local grid manipulations and building-level semantically aware manipulations.