These images are computed from mathematical formulas that assign a color to each pixel. Such a formula is actually an operator network (a directed acyclic Graph) that gets as input the position of the pixel (as a complex number) and a bunch of constants. The outputs of the last three nodes in a topological order of the graph determine the red, green, and blue channels of the color.
The images are bred with this program by selecting the best ones and making variants from them, where the structure of the formula and/or the values of the constants have been changed randomly. Thus, under guidance of the user, an inconspicuous gray blur develops into ever more complex images.