Architectures · Established · Advanced

Generative Adversarial Network

Also known as: GAN

Two networks trained against each other — a generator producing samples and a discriminator judging whether they are real.

What Generative Adversarial Network is

GANs framed generation as a game. The generator improves by fooling the discriminator, and the discriminator improves by catching fakes, pushing both towards realism.

How it works

Both networks train simultaneously with opposing objectives. Instability, mode collapse and careful balancing are the practical challenges, addressed by architectural and loss variants.

Why it matters

GANs produced the first genuinely photoreal synthetic faces and dominated image generation before diffusion models took over. They remain useful where fast single-step generation matters.

Common uses

→Super-resolution and restoration
→Fast image and voice synthesis
→Data augmentation
→Style transfer

Strengths

✓Very fast single-pass generation
✓Sharp outputs

Watch for

✓Unstable training
✓Mode collapse reduces diversity

Continue exploring

Diffusion Model Deepfake AI Image Generation Autoencoder