Recently, there has been growing interest and various discussions in the community regarding the use of generative AI technologies used in inZOI. To provide clear and accurate information, we would like to share the following details.
[Text to Image (T2I) Technology]
The Text to Image (T2I) technology applied in inZOI was trained using publicly available images that are permitted for commercial use.
Over the course of several months, we built our own dataset by selecting 20 million images from public images released under Creative Commons licenses permitting commercial use and modification.
In this process, we referred to MOSAICML’s Common Canvas methodology and applied additional filtering standards to enhance reliability.
For model training we used open-source AI architectures licensed for commercial use.
[3D Printing AI Technology]
Our 3D printing AI technology was trained on approximately 46,000 high-quality 3D models either owned by us or sourced from datasets that are also permitted for commercial use.
This enables users to automatically generate blueprints for in-game 3D printing.
[Video-to-Motion Technology]
The Video-to-Motion(V2M) system was trained primarily on over 1.7 million motion data samples owned and collected by us supplemented with a limited number of publicly available datasets released under open licenses.
As a result, the V2M system enables characters to naturally perform human-like movements such as walking, greeting others, and dancing.
[Our Development Principles]
As we developed and applied generative AI technologies, we generally employed the following principles:
- Use of open-source technologies licensed for commercial use
- Use of carefully selected datasets permitted for commercial use
- Operation of a robust reporting and review system to address any potential claims swiftly and responsibly
We wanted to share this for transparency, as part of our effort to deliver technologies that creators trust.
Thank you,
The inZOI Team
