> ## Content Index
> Fetch the complete content index at: https://itsfoss.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Roblox Unveils Cube 3D: An Open Source AI Model for 3D Creation
- URL: https://itsfoss.com/news/roblox-cube-3d/
- Published: 2025-03-18T11:08:00.000Z
- Updated: 2025-03-18T11:08:00.000Z
- Description: Cube 3D is Roblox's bet on revolutionizing 3D content creation.
- Author: Sourav Rudra
- Tags: Open Source, News, #Import 2025-11-08 21:15

[Roblox](https://www.roblox.com/?ref=itsfoss.com) is a popular game platform that has a passionate community of gamers behind it. The platform lets them make, share, and play games, while also giving them a way to make money with tools like [Roblox Studio](https://create.roblox.com/?ref=itsfoss.com).

Six months after teasing an [open source AI 3D world creator](https://itsfoss.com/roblox-open-source-ai/), they have now open sourced the first release of the foundational model that makes it all happen.

## Roblox Cube 3D: What to Expect?

Named Cube 3D, this [foundational AI model](https://en.wikipedia.org/wiki/Foundation%5Fmodel?ref=itsfoss.com) can be used to generate 3D objects and environments from simple text prompts. Roblox has showcased a demo (*the video above*) that shows it in action, effortlessly creating new 3D elements according to prompts.

The basis for this model is **a novel 3D tokenization technique** spearheaded by Roblox, where 3D objects are represented as tokens. With this approach, complex 3D structures are broken down into smaller, more manageable components, similar to how words are [tokenized](https://en.wikipedia.org/wiki/Lexical%5Fanalysis?ref=news.itsfoss.com) in natural language processing.

Using these tokens, the model predicts the next '*shape token*' in sequence to construct complete 3D objects or even entire scenes. This approach allows Cube 3D to quickly generate functional and structurally accurate 3D models that are fully compatible with game engines.

Roblox describes the 3D generation process like this:

> To achieve 3D generation, we designed a unified architecture for autoregressive generation of single object, shape completion, and multiobject/scene layout generation. Autoregressive transformers are neural networks that use previous inputs to predict the next component. 

> This architecture provides both scalability and multimodal compatibility so that as we expand the model, it will work with many different kinds of input (text, visual, audio, and 3D).

The source for Cube 3D is set to appear on Roblox's [GitHub](https://github.com/Roblox?ref=itsfoss.com) page later this week, if you are interested in learning more about the inner workings of the model, I highly suggest you go through the [announcement blog](https://corp.roblox.com/newsroom/2025/03/introducing-roblox-cube?ref=itsfoss.com).

**Suggested Read** 📖

[CrossOver 25 Unlocks More Windows Games for Linux and macOS GamersCrossOver 25 lets Linux and Mac gamers play an even wider selection of Windows games.![](https://itsfoss.com/content/images/icon/android-chrome-192x192-35.png)It's FOSS NewsSourav Rudra![](https://itsfoss.com/content/images/thumbnail/crossover-25-release.png)](https://itsfoss.com/crossover-25-release/)