Roblox introduces Cube 3D, a GenAI-powered tool for generating 3D objects and scenes, streamlining game development.
Roblox has launched Cube 3D, its first open-source GenAI model for 3D object and scene generation. Built to solve the challenge of limited 3D training data, Cube 3D allows developers to create complex, fully functional assets from simple text prompts. Unlike traditional methods that rely on 2D images for reconstruction, Roblox leverages its vast native 3D dataset to train Cube. Making it highly accurate for in-game assets.
The AI model tokenizes 3D objects similar to how language models process words. Allowing it to predict and generate shape structures efficiently. This method ensures that assets are not just visually correct but also structurally sound, ready for gameplay. Future iterations will expand Cube 3D into full scene generation, where the model predicts layouts and fills environments with detailed objects dynamically. By open-sourcing Cube, Roblox invites developers to fine-tune and extend the model. Fostering broader adoption in the gaming and AI communities.
Developers using Cube 3D gain a powerful tool to accelerate content creation. The AI-driven Mesh Generation API allows quick prototyping—developers can type “/generate a motorcycle,” and within seconds, a fully realized 3D model appears. This dramatically reduces the time spent modeling basic assets, letting creators focus on gameplay mechanics and world-building. Roblox envisions a future where Cube enables players to generate and customize in-game objects, unlocking unprecedented creative freedom.
Beyond creation, Cube 3D represents a step toward AI-driven immersive environments. As the model evolves, it will support multimodal inputs like text, images, and video, moving toward interactive 4D content where objects not only exist but respond dynamically to their surroundings. With Cube 3D, Roblox is setting the foundation for AI-powered game development at scale.