Recently, Roblox, a world-renowned metaverse platform, announced the launch and open source of its first basic AI model for generating 3D objects - Cube3D. This innovative move marks a major breakthrough for Roblox in the field of AI technology, aiming to provide developers with more efficient 3D creation tools and promote innovation in metacosmic content creation.
Roblox official website
What makes Cube3D models unique is its innovative training methods. The Roblox team draws inspiration from text prediction models and develops a technology that can tokenize 3D objects. Through this technology, Cube3D can predict the next shape mark, thereby gradually building a complete 3D object. This method not only improves the efficiency of 3D modeling, but also provides developers with more flexible creative space.
Regarding the training data, Roblox spokesman Samantha Spearman said the model uses a "combination of licensed and publicly available data sets, as well as empirical data from the Roblox ecosystem." This diverse data source ensures wide applicability and high-quality output of Cube3D models.
Roblox also revealed Cube3D's future development plan, saying that the model will eventually become a multimodal model that can accept multiple input types such as text, images, videos, and will integrate with Roblox's existing AI creation tools. It is particularly noteworthy that future versions will be able to use images as input to generate 3D objects, which will greatly expand its application scenarios and creative possibilities.