Official Implementation of MultiWorld: Scalable Multi-Agent Multi-View Video World Models
-
Updated
May 12, 2026 - Python
Official Implementation of MultiWorld: Scalable Multi-Agent Multi-View Video World Models
Action-conditioned game world model: fine-tunes HunyuanVideo (8.3B DiT) with LoRA on NitroGen gameplay data to generate future video frames from gamepad action sequences. Single-GPU training via NF4 quantization + flow matching.
A decision-oriented benchmark framework for evaluating action-conditioned world models beyond static AI benchmarks.
Add a description, image, and links to the action-conditioned topic page so that developers can more easily learn about it.
To associate your repository with the action-conditioned topic, visit your repo's landing page and select "manage topics."