Roblox Corporation has released its foundational 3D model Cube as open source software, making it available to all developers as part of its first wave of generative AI tools. This move allows developers to customize the model through fine-tuning, plugin development, or training on their own datasets.
Core Features of Cube
The Cube model generates fully functional 3D objects and will serve as the foundation for Roblox’s future AI tools. By making it open source, Roblox aims to support researchers, developers, and the AI community in advancing 3D generation technology across the gaming industry.
The model works by tokenizing 3D objects and understanding shapes as tokens, then predicting the next shape token to build complete 3D objects. Training data for the model includes a mix of licensed and publicly available datasets, along with experience data from the Roblox ecosystem.
New Creation Tools
Alongside Cube, Roblox has unveiled several companion AI tools:
- Mesh Generation API: Integrated within Roblox Studio, this tool allows developers to create objects through text prompts, reducing time-consuming mesh generation steps and enabling rapid idea testing without significant resource investment.
- Text Generation API: Enables text-based AI features such as narration and voice commands, creating prompt-and-response interactions between players and the open-source language model.
Player and Developer Advantages
For players, these tools offer expanded creative possibilities, making gameplay more customizable and interactive. Users can create their own items within experiences, which Roblox notes will increase engagement and benefit developers.
Roadmap and Enhancement Plans
Roblox has outlined plans for additional AI tools focused on complex scene generation. The company intends to evolve Cube into a multimodal model trained on text, images, and video inputs that will integrate with its existing AI tools.
The company also plans to enhance Cube to generate objects using images as inputs in future iterations.
