Kakao Brain Announces Paper on Text-Based Motion Generation Model ‘FLAME’

The text-based motion generation model 'FLAME' jointly researched and developed by Kakao Brain and Professor Choi Sung-jun of Korea University. (Photo by Kakao Brain)

[Asia Economy Reporter Seungjin Lee] Kakao Brain announced the text-based motion generation model ‘FLAME,’ co-developed with Professor Seongjun Choi’s research lab at Korea University, at the world’s most prestigious artificial intelligence conference ‘AAAI 2023.’ This research is a new model that can generate human motion using a super-large artificial intelligence (AI) system.

‘AAAI’ is the world’s top-level AI conference hosted by the Association for the Advancement of Artificial Intelligence and was held on the 7th in Washington DC, USA. The Kakao Brain-Korea University Professor Seongjun Choi research lab submitted a research paper on the model ‘FLAME,’ which generates natural motions by inputting desired actions in text form and even modifies motions according to preferences. The paper received high praise at the conference and was selected for an oral presentation.

‘FLAME’ is a motion generation model developed based on a ‘large language model (LLM)’ and is characterized by producing high-quality motions even when detailed and complex sentences are input, compared to the ‘Motion Diffusion Model (MDM)’ announced around the same time.

Professor Seongjun Choi of Korea University said, “It was great to collaborate with the Kakao Brain team, which has shown outstanding achievements in the domestic generation model field, and I am satisfied that the research results received high evaluation. While many companies are focusing on text-based image or video generation, we will continue research on text-based motion generation to create differentiated research outcomes.”

Ildoo Kim, CEO of Kakao Brain, said, “We are honored to reveal the excellent research results of ‘FLAME’ at the global AI conference ‘AAAI.’ If the ‘FLAME’ technology is introduced into areas that were only possible through manual work via motion capture in the production processes of games, animations, and movies, it will greatly contribute not only to automation but also to improving work efficiency.”