In this work Mario Klingemann revisits his "Hyperdimensional Attractions" algorithm in which neural feature vectors become attractors in a high-dimensional 3-body problem inside the latent space of BigGAN. The location of each feature vector translates into a unique image, the closer the vectors pass each other in their virtual orbits the more similar the resulting images they represent will look. In "Hibernation" the starting vectors have been carefully chosen to be almost but not exactly similar, resulting in a calm play with subtlety and creating a counterpoint to the typical fast and loud GAN animations.

AI generated video loop / 4K / 1 minute