With the 2025 Game Developers Conference underway in San Francisco, the online game platform and creation system Roblox has launched its Roblox Cube 3D generative AI model for building 3D objects and scenes.
The open-source foundational model can generate 3D models and environments from text prompts for now, but there are plans to add support for reference images. Ultimately, Roblox plans to expand the model to allow what it calls '4D creation' compatible with the best game development software.
Anyone can use Cube 3D, and it's not necessary to use the the Roblox platform. Since it's being made open source, anyone can fine-tune, develop plug-ins for, or train Cube 3D on their own data. Roblox says the AI allows developers to model props or design their space much faster, with no need to spend hours modeling simple objects.
In the announcement, Roblox notes that current state-of-the-art 3D generation uses images and a reconstruction approach to build 3D objects, which it says is a good option when there isn’t sufficient 3D training data. But Roblox trains on native 3D data. The generated object is fully compatible with game engines today and can be extended to make objects functional, it says.
It compares the difference to a racetrack movie set. "On TV, you might see what looks like a fully functional racetrack, with stands, garages, and a victory lane. But if you were to walk around on that set, you’d quickly realize that the structures were actually flat. Building a truly immersive 3D world requires complete, functional structures, with garages you can drive into, stands you can sit in, and a victory lane with a functional podium.
"To achieve this, we’ve taken inspiration from state-of-the-art models trained on text tokens (or sets of characters) so they can predict the next token to form a sentence. Our innovation builds on the same core idea. We’ve built the ability to tokenize 3D objects and understand shapes as tokens and trained Cube 3D to predict the next shape token to build a complete 3D object. When we extend this to full scene generation, Cube 3D then predicts the layout and recursively predicts the shape to complete that layout."
Roblox said the core technical breakthrough was 3D tokenization, which allows it to represent 3D objects as tokens in the same way that text can be represented as tokens. This gives us the ability to predict the next shape just as language models predict the next word in a sentence.
To achieve 3D generation, Roblox designed a unified architecture for autoregressive generation of single object, shape completion, and multiobject/scene layout generation. This architecture provides scalability and multimodal compatibility so that as Roblox expands the model, it will work with different kinds of input.
Roblox says Cube will underpin more AI tools in the years to come, including highly complex scene-generation tools. The plan is to make it a multimodal model, trained on text, images, video and other types of input, integrating with the company's existing AI creation tools through its AI-powered assistant in Roblox Studio.
Ultimately, the objective is for the 3D objects and scenes generated to be fully functional. Roblox calls this '4D creation', the fourth dimension being the interaction between objects, environments and people. Achieving this requires the AI's ability to to understand the contexts and relationships between those objects.
The news provides another example of how AI is changing game design. Next week, Nvidia ACE-powered AI-generated characters will make their debuts in two games: Naraka: Bladepoint Mobile and the Sims-like inZOI. However, the latest Unity Gaming Report suggests that AI isn't the 'creative cure-all' that some hoped.
