Specspace participates in the "2025 Broadcast Video AI Learning Data Construction Project."

Artificial intelligence (AI) company Specspace (CEO Minho Yeom) announced on the 25th that it will participate in the '2025 Broadcast Video AI Learning Data Construction Project' promoted by the Ministry of Science and ICT and the Korea Radio Promotion Association (RAPA).

This project aims to develop a Korean-style AI model for the production and utilization of broadcast and video content, and to support the digital transformation of the domestic media industry by building diverse content-based learning data.

For this project, Specspace plans to collaborate with domestic broadcasting and cultural institutions, including KBS, KBS MediaTek, Naver Cloud, and the National Heritage Promotion Agency, to build a high-quality AI learning data volume of over 5,000 hours. Specifically, this will consist of ▲ a "video description-based VLM dataset" (4,500 hours) that can automatically explain the meaning of video content, ▲ a "broadcast directing framing-based dataset" (500 hours) that reflects the know-how of professional directors, and ▲ an "asset dataset" (1,500 cases) for creating 3D backgrounds and objects.

Specspace oversees the processing and operation of the entire dataset, particularly in the 3D asset construction and verification sector. Leveraging its AI-based content creation technology, it has built a dataset capable of learning the visual elements of key K-content genres, such as historical dramas and period dramas. This is expected to contribute to increased efficiency in broadcasting and video production.

The data generated through this project will be used to train AI models for domestic broadcasting and media companies, and some of it will be made public through the AI Hub, contributing to enhancing technological competitiveness across the industry.

Specspace CEO Minho Yeom said, “This project is meaningful in that it builds AI learning data that reflects the sentiment of Korean culture and, through this, supports the spread of competitive K-content around the world.” He added, “We plan to continue to work on creating a data-based ecosystem that combines technology and culture.”

• See more related articles