Xindi Yang

[Xindi Yang @Madaoer_Yxd]

Great work! An agent that keeps the structure + a video model that renders the pixels is a strong recipe for world models.
In Oneira we made a similar bet: a coding agent maintains an explicit world state, so interactions persist over long horizons.
More: https://t.co/DhvJxnZb2j

[Google @Google]

Today we're introducing Genie 4 — our newest interactive world model from @GoogleDeepMind 🌐

[Xindi Yang @Madaoer_Yxd]

Key ideas:
- An extensible world state table for interaction with generated content
- 3D boxes in world coordinates for instance-level interaction
- State updates in the table for long-horizon memory

[Xindi Yang @Madaoer_Yxd]

🌍 Introducing Oneira, an interactive video world model where a coding agent keeps an explicit world state.

Objects that appear as you explore become interactive, and your changes persist over long horizons.

Project: https://t.co/DhvJxnZb2j
Paper: https://t.co/7e4uo2Iqu2

[Xindi Yang @Madaoer_Yxd]

Video world models generate open-ended worlds, but their state stays implicit in latent context. Game engines and simulators make every object's state explicit, though only in a prebuilt world.

Oneira keeps an explicit world state that grows with the generated world.

[Baolu Li @calm_lu_312]

📢 VLIPP:Towards Physically Plausible Video Generation With Vision and Language Informed Physical Prior 📢
We propose a two-stage approach that integrates physics as condition into VDM to align the generation with physical principles.
🔥Project_Page：https://t.co/xvm7qkiXRt