project page: arxiv: code:

💥 Probabilistic physics engine: Captures distributions of complex dynamics (collisions, lighting effects, hair simulation, etc.), fundamentally differentiating it from reconstruction-rerendering paradigms.

🏔️ 4D representation: Models 4D distributions to approximate real-world states, achieving superior temporal consistency compared to 2D image-based approaches. Explicit global geometry modeling enables natural integration of spatial memory for long-term coherence.

📃 Text-guided control: Uses text as a universal control modality during training, mapping actions to text for robust, high-frequency, and consistent control during inference.

It demonstrates exceptional capabilities in modeling dynamics, geometry, and physical laws, establishing itself as a 4D world model through three key features:.

Can you imagine playing various games through an AI model? Like BlackMyth: Wukong.🤩.Sharing our latest work: DeepVerse, an autoregressive paradigm-based world model🌏DeepVerse can fantasize the entire world behind images and enable free exploration through interaction🎮.

RT @HaoyiZhu: 🚀 Introducing Aether: Geometric-Aware Unified World Modeling! 🌍. Aether jointly optimizes three core capabilities:.🔹 4D Dynam….