GAN (Generative Adversarial Network) is a type of AI made of two neural networks that compete: one generates fake data, the other tries to spot the fakes. Through this contest, the generator learns to produce increasingly realistic output.
What it means in plain English
A GAN pits two networks against each other. The “generator” tries to create convincing fake data (say, realistic faces); the “discriminator” tries to tell real from fake. As they train, the generator gets better at fooling the discriminator, and the discriminator gets better at catching it. This adversarial back-and-forth pushes the generator toward strikingly realistic results.
GANs were a landmark in generative AI, though for image generation they have largely been overtaken by diffusion models.
A simple example
Many early “this person does not exist” websites, showing photorealistic faces of people who are entirely fabricated, were powered by GANs — the generator having learned to produce convincing human faces through its contest with the discriminator.
Why it matters
GANs were a hugely influential idea that showed how AI could generate realistic, original content. While newer methods now lead for many tasks, the adversarial concept remains an important part of the history and toolkit of generative AI.
Related terms
- Diffusion Model — the approach that largely succeeded GANs for images.
- Generative AI — the field GANs helped pioneer.
- Deepfake — realistic fake media, historically often GAN-based.
Frequently asked questions
What is a GAN?
A Generative Adversarial Network pits two neural networks against each other — a generator that creates fake data and a discriminator that tries to spot fakes — improving the generator until its output looks real.
What are GANs used for?
They have been used to generate realistic images, faces, art, and data augmentation, though diffusion models now lead much of AI image generation.
