Dexterous World Models (DWM)
Seoul National University / VCLab
Scene- and hand-motion-conditioned video generation for studying interactions in static 3D scenes.
What you can explore
Explore the documented training and inference workflow, scene consistency and hand-conditioned prediction.
Before you use it
Video generation is not a robot action policy or a validated contact simulator. Confirm checkpoint availability, preprocessing and GPU requirements before attempting reproduction.
Original contributors
Byungjun Kim, Taeksoo Kim, Junyoung Lee and Hanbyul Joo
License / access: Code: Apache 2.0; upstream models and datasets have separate terms
Official implementation & setup
Go deeper into this methodInputs, outputs, evidence, compute and release limits →
Editorial reference · checked 2026-09-23