1 briefs tagged #kvae.
The arXiv report describes audio, image and video tokenizers for text-conditioned generation.