One framework, Three directions
Act in different worlds.
Action-conditioned image-to-video generation
Same action across worlds.
Action-conditioned text-to-video generation
Action + text → video
Results unavailable.
Loading scenes…
Shared input Generated video Generated video Generated video
Motion reference
Recover 3D motion from video.
Egocentric pose estimation
Results unavailable.
EgoDex
Imagine worlds with your words.
Text-only joint action–video generation
Generated video
Generated action
Loading video and action…
Front
Side
Top
Act and imagine.
World-action policy
01 Choose a scene
02 Choose a target
Initial observation
Scene unavailable. Please select another example.
HOT3D
More applications
Human-to-robot and background transfer
Overview