Agenda

The Hidden Ones, a 3D fighting game by Tencent Games, showcases the beauty and precision of Kung Fu through lifelike character motion. Traditional motion generation methods often suffer from visual artefacts such as foot sliding and rigid transitions. Previous AI solutions failed to meet the demands of real-time applications due to limitations in model size, inference speed, and post-processing complexity. This talk introduces a novel real-time AI technique for generating fluid Kung Fu motion transitions, designed specifically for the demands of fighting games. The solution combines marker-less motion capture for data collection, a dedicated generation model, and streamlined post-processing for deployment. Compared to conventional approaches, it significantly reduces data acquisition costs, shrinks package size, speeds up inference, and generates smoother, more natural transitions—especially in nuanced movements like wrists and feet. Already implemented in The Hidden Ones, this technique elevates visual realism and Kung Fu motion fluidity in fast-paced, real-time combat scenarios.

Takeaway

Attendees will learn about the challenges of applying real-time Fung Fu motion generation in 3D fighting games and a novel solution to apply game animation AI to reduce resource consumption while ensuring motion quality.

Intended Audience

This lecture is for AI developers and those interested in Gen AI and animation system including animation generation. Having basic knowledge of game animation system and knowledge about ai application would be helpful but is not required. All interested attendees are welcome.