V-JEPA by Meta

« A non-generative LLM model that learns by watching videos. It produces excellent recognition and detection results »