VibeWorlding introduces a multimodal agent framework for autonomous 3D scene layout and tool invocation from user queries. It includes VWE-BENCH, containing 2,616 3D assets, 323 human-annotated seed worlds, and 6,828 synthesized multimodal queries.
HOW THIS AFFECTS YOU
●
builderYou can leverage this framework to build agents capable of end-to-end 3D world generation.
●
researcherYou can use this benchmark to evaluate how agents reason over multimodal 3D information.