Alibaba-owned Gaode has launched ABot-WorldStudio, a world model workshop that generates interactive 3D worlds from text or image inputs, featuring a 'spatiotemporal any door' for teleporting between worlds and supporting unlimited-duration local inference on a single RTX 5090. This release marks a significant advancement in world model technology by enabling long-duration local inference and unifying interactive video generation with 3D Gaussian Splatting (3DGS) scene generation, potentially accelerating applications in embodied AI, game development, and virtual tourism. ABot-WorldStudio can generate interactive 3D worlds as both video and 3DGS files, and the underlying ABot-World models are fully open-sourced. In tests, it sustained continuous inference for over one hour without crashes or quality degradation, far surpassing the typical one-minute limit of similar products.
Background
World models are AI systems that build internal representations of environments and predict how they change over time, enabling agents to plan and act without real-world trial and error. 3D Gaussian Splatting (3DGS) is a volume rendering technique for high-quality, real-time 3D scene representation from multiple images. ABot-WorldStudio combines these concepts, allowing users to create and navigate immersive 3D worlds locally.