PAN 2 Is An Open World Model That Predicts What Happens Next, Not The Next Word
0:00 / 0:44
News
PAN 2 Is An Open World Model That Predicts What Happens Next, Not The Next Word
calendar_today Date:
schedule Duration: 0:44
database
Summary Report
MBZUAI's IFM releases PAN 2, an open 29B-parameter world model that generates video of what happens after an action.
- 01. Predicts the next world state from a scene and an action; up to 20s at 24fps
- 02. 29B diffusion model; weights and training code under Apache 2.0
The Institute of Foundation Models at the Mohamed bin Zayed University of Artificial Intelligence has released PAN 2, an open world model. Where a language model predicts the next word, PAN 2 predicts the next world: describe a scene and an action, and it generates video of what happens next, grounded in time and physics, for up to 20 seconds at 24 frames per second. A language model writes out the next state of the world, and a 29-billion-parameter diffusion model renders it. Weights and training code are open under Apache 2.0. IFM notes that in roughly two of every three long plans, at least one step isn't carried out.
Meta Data
Company:
Model:
