SCOPE: Simulating Cross-game Operations in Playable Environments for FPS World Models
TL;DR AI
2 min readKey summary
Researchers introduced SCOPE, a video diffusion method for FPS world models that localizes action effects while keeping unaffected parts of the scene stable.
SCOPE conditions pretrained video diffusion transformers per pixel to separate weapon-related changes from broader scene updates, without needing segmentation labels.
The paper also releases CrossFPS, a 69K-clip dataset from 7 FPS titles with aligned 10-DoF action data.
Together, these advances improve interactive game simulation and support cross-game transfer without game-specific training.
