This is a live demo of the generator ported from the Tensorflow Generative Adversarial Network model trained on the Fashion-MNIST dataset to deeplearnjs. You can find more information about it in the release post and source codes on github.
Basically, this network takes some numbers as inputs and produces an image. Generate random button assigns random values to inputs and generates an image from them. Generate batch does exactly the same thing 16 times. Generate from inputs produces an image based on input control values. You can tweak them as you like to see how they affect final results. When you will get too much of them, Clear results will help you to start over.