NC AI, the artificial intelligence subsidiary of NCSoft, said Monday it has demonstrated its “World Foundation Model,” highlighting its research infrastructure and positioning the technology as a potential “K-World Model” for Korea’s AI ecosystem.
The World Foundation Model, or WFM, uses a simplified architecture that generates actions directly from latent space information — data processed before images are produced — improving both efficiency and prediction accuracy.
The model also addresses the so-called “Sim2Real gap,” a common problem in robotics where systems trained in simulation fail to perform reliably in real-world environments. According to the company, the WFM showed strong predictive capabilities and stable performance when applied to robotic tasks.
The research also demonstrated significant improvements in resource efficiency. WFM training required only about 25 percent of the GPU resources typically used by leading models while maintaining high success rates in robotics tasks linked to potential commercial applications.
To address data shortages in industrial AI training, the system can generate large volumes of synthetic video data. NC AI said the WFM environment can produce roughly 10,000 hours of synthetic video within 11 days through simple prompt-based operations, reducing the need for costly real-world data collection.
The company plans to supply domain-specific synthetic data tailored to Korean manufacturing industries, including semiconductor cleanroom operations, steel production processes and shipyard block assembly.
“With our world model technology, we will work with the K-Physical AI Alliance to build a robotics ecosystem specialized for Korean industries,” said Lee Yeon-soo, chief executive of NC AI.
NC AI is a core member of the K-Physical AI Alliance, a consortium of companies, government-funded research institutes and universities aimed at advancing AI-driven technologies by leveraging Korea’s strong manufacturing infrastructure.
stlee0329@heraldcorp.com
