A Variational Autoencoder is a neural network architecture that can be trained on a batch of similar looking images to generate new, but similar looking images based on the characteristics it understands of the batch. I trained a Variational Autoencoder on a sequence of images of pixels filling in a square - in other words, two opposite images and every stage of difference in between.
A total of 8 different transition sequences were tested. Horizontal left to right, horizontal right to left, vertical left to right, vertical right to left, and each of their inverses. My stupid human brain kind of secretly hoped the result would make sense. Like a simple average of the images or something like that. I ended up with some weird gradients. These were then 3D printed into puzzle pieces that do not quite fit.