UniVerse3D: Emerging Properties of Unified Multimodal Models in 3D Understanding and Generation

While unified multimodal models have made progress in the 2D domain, extending this paradigm to 3D remains in its nascent stages, hindered by data scarcity, high generative complexity, and the lack of well-aligned semantic representations. To address these challenges,