Introduction
An ordinary GAN generates random samples: you can't ask it for "a 7" or "a cat". A conditional GAN takes an extra input, a condition, that tells it what to generate. The condition can be a class label, a text description, or even a whole image.
When the condition is an image, we get image-to-image translation: turning a sketch into a photo, a map into a satellite view, a daytime scene into night, a horse into a zebra. Think of a skilled illustrator who's handed a rough pencil outline and asked to produce a finished, coloured painting that keeps the same composition. The outline is the condition; the illustrator is the generator.