StyleGAN2
lastmod 2026-11-10created 2022-06-094 edits
StyleGAN2 is the successor to StyleGAN, a generative adversarial network (GAN) released by NVIDIA, proposed in the paper "Analyzing and Improving the Image Quality of StyleGAN". It removes the droplet-shaped artifacts that appeared in original StyleGAN images by replacing AdaIN normalization with weight demodulation, and introduces path length regularization and a network architecture without progressive growing to improve image quality and the smoothness of the latent space. It is a representative baseline model for high-resolution image synthesis such as face generation.
Analyzing and Improving the Image Quality of StyleGAN