First video tests with the stylegan2-ADA GAN network.
Stylegan2-ADA is a generative adversarial network machine learning framework that uses adaptive discriminator augmentation (ADA) allowing for training with small datasets of images.
Using Google Colab, this model was trained on frames from a video of a train ride through Italy. Moving through the resulting latent space, the network generates it's own interpretation of the original train ride.
The final video includes an original piano composition recorded on Fender Rhodes inspired by the music of Hiroshi Yoshimura.