Generates various portraits in a grid. Trained on 100 3x3 images from 4chan's /vt/ board, with the data based primarily around female vtubers. For best results, go for 768 x 768 square. At 512 by 512 it may generate only 4 images, at 1024 x 1024 it may generate up to 16.
Still requires further testing, so give it a try!
