NVIDIA world-action model jointly predicting video and robot actions.
Access and availability
- Code
- Public access
- Model files
- Public access
Code and training / fine-tuning
- Access
- Public access
Verified on · 2026-10-27
- Code license
- Apache-2.0
Model files
| File | Access |
|---|---|
| dreamzero-agibot | Access |
Getting started
- Deployment requirements
- Official setup tests GB200/H100 GPUs and CUDA 12.9+. Distributed inference requires at least two GPUs; this is not a single consumer-GPU setup.
At a glance
- Model type
- World-action model
- Backbone
- Wan2.1-I2V-14B-480P video backbone with joint action prediction
Technical details
Functional purpose
Action & control / World prediction & simulation
World-model role
Video and actions are jointly predicted during inference.
Before you use this version: Available weights or code do not establish plug-and-play robot compatibility. Check the required configuration, data and interfaces in the source documentation.
Submit a source or correction