What is GAN (Generative Adversarial Network)?

How it works

GANs were the dominant generative AI architecture before diffusion models. A generator network creates content while a discriminator network tries to distinguish generated content from real content. This adversarial training process pushes the generator to produce increasingly realistic output. While diffusion models have largely superseded GANs for image and video generation, GAN-based techniques remain relevant in specific applications like super-resolution, face generation, and real-time style transfer. Understanding GANs provides historical context for the field and helps explain some current tools that still use GAN-based components in their pipelines.