Roblox has unveiled Cube 3D, a foundational model for generative AI that advances 3D content creation on its platform. As part of this release, Roblox is open-sourcing a version of Cube 3D and launching a beta mesh generation API, which will be accessible in Roblox Studio and as an in-experience Lua API later this week.

Cube 3D is a generative AI system that creates 3D models and environments directly from text inputs. In the future, it will support additional inputs like images. Unlike traditional 3D generation methods that rely on image reconstruction, Cube 3D is trained on native 3D data from the Roblox platform, ensuring that its generated models are structurally complete and fully compatible with modern game engines.

This approach differs from traditional 3D rendering methods, much like how a real, fully interactive racetrack is different from a flat set in a movie production. Cube 3D aims to build genuinely immersive 3D environments, where objects are complete and functional in virtual spaces.

How Cube 3D Works

Inspired by AI language models, which predict the next word in a sentence, Cube 3D tokenizes 3D objects, treating shapes as tokens. It then predicts the next shape to construct complete 3D objects. This technique extends to entire scene generation, allowing Cube 3D to predict layouts and recursively generate detailed environments.

The model utilizes a cross-attention mechanism between 3D shape tokens and text/image inputs, making it adaptable to multimodal inputs. This architecture lays the foundation for future capabilities, including scene understanding and interactive world-building.

Open-Source Commitment and AI Community Impact

By open-sourcing a version of Cube 3D, Roblox is inviting researchers, developers, and AI enthusiasts to experiment, fine-tune, and extend its capabilities. This move aligns with Roblox’s broader commitment to AI safety and transparency, which includes contributions to open-source AI safety tools and its involvement in ROOST, a nonprofit dedicated to digital safety.

The Vision for 4D Creation

While Cube 3D currently focuses on single-object generation and mesh completion, it is designed to evolve into a fully multimodal AI model capable of integrating text, images, video, and other inputs. Ultimately, Roblox aims for 4D creation, where the fourth dimension represents interactivity, allowing AI-generated objects and environments to react dynamically to users and in-game conditions.

Roblox is committed to enhancing Cube’s capabilities over time, pushing the boundaries of AI-powered creativity. This isn’t just a concept, it’s a real, functional tool with practical applications for today’s 3D developers and players.