V-JEPA

Meta · 15 February 2024

Video JEPA learns by predicting masked spatio-temporal regions in latent space from unlabeled video; frozen encoder reused across tasks.

Extended I-JEPA to video with 1.5x-6x better training/sample efficiency than generative pixel prediction, as a step toward LeCun's world-model agenda.

Date

Thursday, 15 February 2024

Lab

Meta

Kind

open-weights

Access

open weights (restricted license)

Figures

Released under CC BY-NC (non-commercial).

Sources

1. ai.meta.com/blog/v-jepa-yann-lecun-ai-model-video-joint-embedding-predictive-architecture/

This record was checked against its sources on 6 October 2026. How we check

Related

• I-JEPA