GAN: Generative adversarial network (GAN) are well-known deep generative models proposed by Ian Goodfellow that could be used for synthesising data. It consists of two components, a generator (G) network that learns the data distribution and generates new examples, and a discriminator (D) network that distinguishes between real and fake examples, i.e., examples generated by G.
Conditional GAN: GANs produce synthetic images by drawing a random vector from latent space. However, we may condition the GAN on additional information, namely, a class label, e.g., label "0" in MNIST. This requires additional input of the label to the G and D networks along with a random vector drawn from latent space.