ML//GAN//StyleGAN

StyleGAN is NVIDIA's GAN architecture that maps a latent code through an intermediate style space and injects style controls throughout the generator.

StyleGAN is NVIDIA's GAN architecture that maps a latent code through an intermediate style space and injects style controls throughout the generator.

Coarser layers tend to control pose and global shape, while finer layers influence texture and local detail. This partial disentanglement made latent-space editing unusually legible.

It powered the website This person does not exist, which showed the world what GANs could do.

The original StyleGAN inherited progressive growing; later StyleGAN versions achieved better quality and stability without relying on that training schedule.

The important contribution was not just photorealistic faces. It was making the generator's internal coordinate system useful enough to manipulate.