World ModelMeta AI (FAIR) · 2025

V-JEPA 2

Self-supervised video joint-embedding predictive world model for understanding and planning.

Summary

V-JEPA 2 is a self-supervised video model trained with a joint-embedding predictive objective in latent space (not pixels). It learns physical-world representations from large-scale video and can be used for prediction and action planning. Open source.

Metadata

Organization: Meta AI (FAIR)
Year: 2025
License: Open source
Objective: Joint-embedding predictive (latent)
Availability: Open weights + code

Relationships

Introduced by

V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning

#self-supervised#jepa#representation