This document summarizes a paper on semantic image synthesis using spatially-adaptive normalization (SPADE). SPADE utilizes semantic segmentation maps to modulate normalization layers, better preserving semantic information compared to other normalization techniques. The proposed SPADE generator takes a segmentation map and noise vector as input and produces a photorealistic image. Experiments show SPADE generates higher quality, more realistic images compared to other semantic image synthesis methods on various datasets based on both automatic and human evaluation metrics.