NC AI and South Korean broadcaster MBC are working on a government-backed project to build AI tools that understand more than who or what appears in a shot.
The consortium, which also includes NHN Cloud and DataMaker, has been selected for South Korea’s AX One-Stop Voucher program, run by the Ministry of Science and ICT and the National IT Industry Promotion Agency. AX here means AI transformation, which is a grand phrase for a practical government aim: give companies support for AI software, cloud infrastructure and data work in one joined-up package.
The planned system is built around what the partners call a media ontology. In plain English, that means a structured map of what is happening inside a programme: people, dialogue, emotion, scene context, locations, actions and how those pieces relate to one another.
That matters because a lot of current video AI is still better at spotting isolated things than understanding why a moment matters in the programme. A tool can transcribe speech, tag a face or identify a car. It is much harder for it to know that a laugh is a reaction shot, a scene is the emotional peak of an episode, or a piece of dialogue matters because it resolves a conflict set up earlier.
MBC’s version of the promise is very post-production specific. The broadcaster says the project is aimed at reducing repetitive preparation work: arranging large volumes of footage, summarising material as text and generating early rough assemblies around key people or scenes. The editor would not be replaced by a magic button; the pitch is that the editor starts from a draft rather than a blank timeline.
The same context layer is also intended to support multilingual dubbing and audio description for visually impaired audiences. Those are sensible targets. Localization and accessibility both depend on more than literal transcription. Timing, tone, character intent and scene meaning all matter, which is exactly where crude automation tends to become obvious.
NC AI is expected to lead the technology work, including the ontology engine, model fine-tuning, SaaS platform and APIs. The company also plans to bring in its VARCO media AI tools, which already cover areas such as translation, voice and 3D asset generation.
The caveat is that this is still a development project. The announcements describe what the partners plan to build, not a tool that has already been tested across messy broadcaster archives, rights-sensitive material, complex edits and professional delivery deadlines.
For post teams, the interesting question is not whether AI can identify a person in a shot. It is whether it can produce useful assemblies without inventing meaning, flattening performance or creating extra checking work. Anyone who has spent time fixing “helpful” automation will understand the difference.