A GAN is a Generalised Adversarial Network, which is a way of organising an AI network for training. Basically two AI’s are pitted against one another in order to accelerate the training. Each AI works to out do the other and create a more and more convincing outcome.

Since the advent of GAN’s in 2014 they have produced some amazing results. Most notable among these are the text and image based AI systems which can synthesise convincing new images and texts.Now they are becoming increasingly available to the public.

A year or so ago I tried making some Deepfakes using an AI running on google cloud. This was running on an expensive GPU setup but was free to use as a research tool (although I wasn’t doing research, I was just having fun).

Now I am playing with Artbreeder which uses two GAN’s to create a variety of art. The illustration here is totally AI generated from a selection of input images and tweeks. The AI can generate realistic or stylistic images of different races and genders.


A collaborative tool for discovering images.

Leave a Reply

Your email address will not be published. Required fields are marked *