Grass Unveils AI-Based 'Video Search'

Presenting New Possibilities for Video Data Search

Grass announced the 'Grass Video Search' through its official X channel today, showcasing an innovative feature that can search the actual content of video frames using AI technology. This is considered a significant advancement that goes beyond existing text and tag-based search methods, potentially solving bottlenecks in large-scale video data processing and curation automation.

Over the past few months, Grass has collected over 1 billion videos to build a vast dataset and implemented scene segmentation through key frame annotation using Visual Language Models (VLM). This allows precise capturing of moments when specific visual elements or concepts appear.

Currently, Grass is processing over 4 billion additional videos, with its index growing daily. This technology provides early access opportunities for video and large-scale data analysis researchers, journalists, and non-profit organizations for AI model training or large-scale video work.

The release of Grass Video Search is expected to accelerate innovation in video data analysis and significantly contribute to the development of related industries.

Choi Joo-hoon joohoon@blockstreet.co.kr