CreateAI联合中科院自动化所推出NeoVerse 4D世界模型

Core Insights - CreateAI has officially launched the NeoVerse 4D world model in collaboration with the Institute of Automation, Chinese Academy of Sciences, which is now available for global developers to review [2][5] - The NeoVerse model integrates diffusion and 4D Gaussian core technologies, trained on 1 million segments of open in-the-wild monocular video, enabling the construction of a general 4D world model in just 30 seconds [2][5] - The model addresses the scalability challenges faced by current 4D world model training solutions, breaking the reliance on expensive multi-view data and creating an infinite iterative loop of "reconstruction + generation" [2][5] Industry Applications - The 4D world model demonstrates significant application potential in digital content creation, game development, autonomous driving simulation, and embodied intelligence [2] - NeoVerse serves as a versatile 4D modeling solution that supports various applications such as 4D reconstruction, precise roaming, multi-view generation, video editing, and 3D point tracking, facilitating the transition of 4D technology from specialized to general use [2][5] - The model's ability to efficiently synthesize large amounts of 4D data from monocular video is expected to accelerate the adoption of 4D technology across multiple industries [5] Technical Advancements - NeoVerse supports powerful distillation LoRAs, allowing for rapid inference in under 30 seconds on a single graphics card, showcasing its strong industrial application potential [5] - The innovative approach of using Feedforward 4D Gaussian reconstruction combined with diffusion generation paves the way for the establishment of 4D world models without the need for costly multi-view data [5]