Greetings, Innovators.
Lately, there has been increasing curiosity and diverse discussions within the community about the generative AI technologies employed in inZOI. To offer precise and transparent insights, we wish to provide the following information.
[Text to Image (T2I) Technology]
The Text to Image (T2I) technology utilized in inZOI was developed using publicly accessible images authorized for commercial purposes.
Over several months, we curated a unique dataset by selecting 20 million images from publicly available sources, all under Creative Commons licenses allowing commercial use and modification.
During this endeavor, we adopted MOSAICML’s Common Canvas method and implemented additional filtering criteria to bolster reliability.
For model training, we relied on open-source AI architectures that are commercially licensed.
[3D Printing AI Technology]
Our 3D printing AI technology was trained with around 46,000 premium 3D models, either owned by us or sourced from datasets approved for commercial use.
This technology enables users to automate the creation of blueprints for 3D printing in-game.
[Video-to-Motion Technology]
The Video-to-Motion (V2M) system was primarily trained on more than 1.7 million motion data samples owned and gathered by us, complemented by a few publicly available datasets released under open licenses.
Consequently, the V2M system allows characters to perform human-like actions such as walking, greeting, and dancing naturally.
[Our Development Principles]
Throughout our development and implementation of generative AI technologies, we adhered to the following core principles:
- Utilization of open-source technologies licensed for commercial use
- Selection of meticulously chosen datasets permitted for commercial use
- Maintenance of a robust system for reporting and reviews to promptly and responsibly address any potential claims
We share this for transparency, as part of our commitment to delivering technologies that earn creators’ trust.
Thank you,
The inZOI Team