brains/v-jepa-2 · model record

V-JEPA 2

Self-supervised joint-embedding predictive video model (up to ~1B-parameter ViT-g encoder) pretrained on over 1 million hours of video. The action-conditioned V-JEPA 2-AC, post-trained on under 62 hours of DROID robot video, enables zero-shot pick-and-place planning on Franka arms.