GAN
Also called Generative Adversarial Network
Two networks in a duel: one fakes images, the other calls out fakes, and both get sharper.
Think of it like
A forger and a detective locked in a room — each pushes the other to get better until the forgeries pass.
Example
The eerily realistic faces on “thispersondoesnotexist” were GAN output, back when GANs ruled image generation.
How it actually works
A generator invents images and a discriminator judges real vs. fake; training pits them against each other until the generator’s fakes fool the judge. It produces crisp results but is notoriously unstable to train and prone to “mode collapse,” where it churns out the same few images. Diffusion models have mostly taken over because they’re steadier.
For product teams
A once-dominant approach now mostly of historical and niche interest for generation.
For engineers
A minimax game between a generator and discriminator, trained to a Nash equilibrium.
Related
- Generative Adversarial Network — The full name spelled out.
- Diffusion Model — The method that largely replaced it.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome