So it's based on a neural network where each layer is only one neuron wide, but gets fed from all inputs plus prior neurons. There are two inputs, ten neurons, and the final three give you your rgb value. Each neuron also has a bias (initial value), then it sums each of the inputs and prior neurons by a weights, and then passes that through an activation function. The activation function in this case is tunable. It is Math.min(255,Math.max(-255,Math.sign(x)*Math.pow(x,k))); Where k is also a weight. Each weight was initiallized with a random value from -1,1. Then for rendering I'm not using the same interpretation of RGB values I plan to use in the long run but wanted to use something that would generate something even with "junk" data from a completely random weights. That is Math.abs(v%1)*255. That helps produce a lot of the periodic features. Once I get rid of that and just interpret the output directly it should be far less periodic.