GAN Lab

GAN Lab is an interactive playground that trains a Generative Adversarial Network on a 2D point distribution in your browser, letting you watch the generator and discriminator co-evolve in real time. A GAN pairs two networks in a minimax game: the generator maps random noise to synthetic samples while the discriminator tries to tell those fakes from real data — each training step makes one smarter, forcing the other to adapt. Use GAN Lab to change the target distribution (grid, ring, Gaussian mixture), adjust the learning rates and network sizes, and observe how quickly the generated points spread to cover the real distribution. Watch for mode collapse — when the generator fixates on a few clusters — and see how slowing the discriminator or tuning the step ratio can recover diversity.