The Broadcasting Media Communication Commission and the Korea Radio Promotion Association (RAPA) said on Aug. 5 they will open some AI training data for broadcast video for education and research through the National Information Society Agency (NIA) AI-Hub secure zone.

The disclosed data total 1,175,800 items of AI training data for broadcast video that reflect Korean culture and sentiment. They are made up of 10 types across 3 areas, including video and image understanding and description, background, character and object generation, and upgrades to the production environment.

The data include 294,906 items of broadcast video image captioning, 217,587 items of video captioning and 37,940 items of K-content video understanding data.

They also include 128,689 items of Korean-style background and object generation data, 96,991 items of Korean-style emotion character generation data and 95,446 items of K-content character persona generation data.

In the production-environment upgrade area, 43,604 items of K-content vision language model (VLM) data, 151,334 items of font generation data and 86,350 items of subtitle generation data will be released.

Demand for AI training data for broadcast video has been high, but use has been constrained by issues such as intellectual property rights, portrait rights and neighbouring rights. Bamingtongwi said it expected the release would enable developers to pursue research on AI models that use broadcast video.

The data can be used in the AI-Hub secure zone. The secure zone is a space designed to allow the use of data in a security-controlled environment, and it can be used free of charge after prior application and approval procedures.

Cheon Ji-hyeon (천지현), director-general of the Broadcasting Media Promotion Bureau at Bamingtongwi, said, "Broadcast video is a core asset of the AI industry that brings together Korean culture and sentiment, as well as advanced production technology and expertise." Cheon said, "We will continue to build and disclose high-quality AI training data so that broadcast video can be used across various industries."