V-JEPA is a Meta model that learns about the world from video by predicting the hidden parts of a clip as abstract features, not pixels.
In plain words
Many video models learn by redrawing missing pixels. V-JEPA, short for Video Joint Embedding Predictive Architecture, hides parts of a video and predicts a compact description of what is missing instead. It trains without labels, text or pretrained image encoders. The first version appeared in February 2024. V-JEPA 2, in June 2025, trained on over a million hours of video, and a version tuned on robot videos could plan simple pick-and-place tasks.
A quick example
Imagine showing V-JEPA a clip of a hand reaching for a cup with part of the frames blanked out. Instead of painting in the missing pixels, it predicts a summary of what happens, such as the hand closing around the cup.
Where you'll meet it
- Research on world models and video understanding.
- Robot planning experiments.
Related
Sources
- PaperRevisiting Feature Prediction for Learning Visual Representations from Video, Bardes et al. (arXiv) · read 29 Sept 2026
- Docsjepa, Meta (GitHub) · read 29 Sept 2026
- PaperV-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning, Assran et al. (arXiv) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.