vae-nat using Noise-As-Targets to train VAEs Making a video out of the stored frames: ffmpeg -framerate 25 -pattern_type glob -i '<images_dir>/*.png' -c:v libx264 -pix_fmt yuv420p <output_name> How to train a NAT model to learn an Image Tutorial coming soon