What is the difference from WonderPlay?

[SMY19999]

https://github.com/kyleleey/WonderPlay

Apart from updating the video diffusion model to real-time, what is the difference between this study and WonderPlay?

[ziyc]

Good question!

Actually, the goal, scene representation, and video model are all different.

We target real-time video generation, while WonderPlay targets image-to-4D scene representation reconstruction/generation. WonderPlay represents the scene with Gaussians while we generate frames directly with video models without such representations. Our models are post-trained and distilled directly for real-time video generation instead of scene representation updating.