COSMO TOKENIZER

Summary from the TLDR AI newsletter

Nvidia has released state-of-the-art image and video tokenizer and compression models. These can be used to train video generation, VLMs, and other multimodal models. Nvidia has released the inference code, a paper, and model weights.