LiveStyler: AI 기반 실시간 멀티 스타일 비디오 변환

In existing live broadcasts or video production, post-editing was essential to achieve various visual effects. Applying single filters or offline editing limited real-time creativity. These post-processing steps caused broadcast delays, increased production complexity, and higher costs, hindering agile production of live content.

LiveStyler utilizes DCT-GAN-based AI segmentation technology to transform faces, objects, and backgrounds into different styles simultaneously in real-time. Operating in the cloud, it can be used directly from a browser or SDK without on-site GPUs, delivering Full HD 30fps or higher performance with end-to-end latency of 0.5 seconds or less, enabling various style transitions during live broadcasts. This is innovative in terms of real-time application of multiple styles and high responsiveness compared to existing competitive solutions.

Creators and broadcasters are the main users, with applications ranging from B2C live streamers to B2B broadcast content producers. It targets the entire media and entertainment industry, which values real-time content production, and can be applied to the metaverse avatar industry in the future. Although it is a Korean startup, it targets the global market, and partnerships with content creation platforms orstreaming service providers (B2B2C)are also expected.

Provided as a cloud service, environmental restrictions are minimal, and users can easily adopt it without specific hardware. From a regulatory perspective, real-time content modification is unlikely to conflict with broadcasting regulations, but ethical considerations such as copyright filter application are necessary. Industrially, demand is expected to grow due to the increase in single-person media and the expansion of the real-time streaming market, and it can be expanded to other real-time video fields such as video conferencing and education. However, restrictions may exist in some countries depending on AI video synthesis regulation trends.

The CES Innovation Awards judges highly praised the technical completeness and innovation. The demonstration of various filters being applied without delay in the actual use demo was evaluated as proof of high-level technology. Market expectations are also positive as a game-changer in the live content market, and innovation was recognized in that post-production was made real-time. As for overestimation factors, there was some caution as to "whether the AI style diversity will fully satisfy the actual broadcasting quality," but the dramatic reduction in post-processing time was highly evaluated.