世界模型NeurIPS 20252025

Genie 3: Interactive 3D World Models from Text

Yixia Li, Hongru Wang, Peng Lai, Zhiwen Ruan, He Zhu, Youxin Zhu, Ganlong Zhao, Minda Hu, Yun Chen, Sibei Yang, Peng Li, Jeff Z. Pan, Jia Pan, Guanhua Chen, Yang Liu, Guanbin Li et al.Google DeepMind

下载 PDF

摘要

We present Genie 3, a generative model that creates interactive 3D environments from text descriptions. Genie 3 produces navigable 3D worlds with consistent physics and object interactions, enabling users to explore generated environments in real-time with full six-degrees-of-freedom control.

3D world generationinteractive environmenttext-to-3Dnavigationreal-time rendering

技术细节

数据集

Text-to-3D DatasetShapeNetObjaverse

仿真平台

PyTorch3D CUDA Renderer

模型骨架

Text-conditioned 3D Generation + Interaction Model

编码器

T5-XXL Text Encoder + 3D Point Encoder

解码器

3D Gaussian Splatting Decoder + Interaction Decoder