Few-shot Semantic Image Synthesis Using StyleGAN Prior Our method can synthesize photorealistic images from dense or sparse semantic annotations using