NeuralKuvshinov version 2

Recently, after making the first NeuralKuvshinov, I found a new way to preprocess the dataset of Kuvshinovs' artwork. I thought, "why not?" and use styleGAN2-ada-pytorch and stylegan2-ffhq-config-f.pkl to make a second version of NeuralKuvshinov.
Boy, it was a good idea. See yourself:

For ones who don't understand the significance of this leap to greatness, this is a common random grid from first NeuralKuvshinov:

and this is from the second:

As usual, here some bonus tiers:

men

megane

scribbles

In order to generate images by yourself, use this colab.

Table of checkpoints:

bonus videos

Interpolation

LucidSonicDream music video