These animations were created by taking photographic composites and chaining them to create generative animations. Most frames a designed without an overall consideration of past/future frames. Each frame is effectively it's own painting. Combining them (even in random ways) results in interesting, "moving pictures" with a surprising degree of coherency. Creation of said images can easily be automated.

​

While it is completely possible to create animations in this way without the use of neural networks-- Using neural networks allows for fast creation of complicated animation frames. Using style transfer (or even just some vanilla processing) on the same frame repeatedly can create the illusion of basically cohesive movement without the need for tweening frames. In principle these types of animations are meant to be rapid-prototyped. They may be used to quickly express an idea or abstract concept without  the need to invest much time in crafting the animation frames. 

​

Photographic composites are created by double exposing film or film sensor or may be created artificially. I've found that using long-exposed photographs in sequence yields interesting results. For certain frames/images a GAN was used to induce style transfer from a reference painting. 

​

Future: Auto-generation of tween frames for seemless transitions and collaborative animations.