SEOUL, July 14, 2025 — Twigfarm, a leading content AI company, announced today that it has been selected for two major projects in the first round of the "2025 Hyper-Scale AI Ecosystem Expansion" initiative, led by the Ministry of Science and ICT and the National Information Society Agency (NIA) of South Korea.
The two awarded projects are:
Both projects are focused on building high-quality multimodal datasets for AI training and enhancing content utilization.
The K-Stock Content Data Construction project aims to create image-text based multimodal datasets centered on traditional Korean culture, cuisine, and nature. Over 1 million bilingual (Korean-English) captions—surpassing 10 million tokens—will be developed and published on the AI Hub, an AI platform operated by NIA that offers data, software, and computing resources essential for AI R&D.
Twigfarm will implement a CoT (Cognition of Thought)-based three-stage labeling system that reflects human reasoning to overcome the limitations of existing stock content, such as bias and lack of contextual explanation. The goal is to produce optimized data for Korean-style generative AI training.
In the Media & Content De-identification and Clean Data Construction project, Twigfarm will contribute its technology to automatically identify and remove sensitive elements such as faces, subtitles, and logos from broadcast videos. This will result in the creation of clean videos and object-caption datasets, maximizing the reusability and value of media content.
Twigfarm has been advancing its proprietary video localization platform, LETR WORKS, which integrates multilingual subtitles, AI dubbing, resolution enhancement, audio cleaning, and automated clean video generation—all powered by cutting-edge AI.
A Twigfarm spokesperson stated,
"Through this project, we aim to build essential multimodal datasets based on Korea’s unique cultural content and contribute to establishing data sovereignty in preparation for the sovereign AI era."