Video game publisher and developer Cygames, Inc. (HQ: Shibuya, Tokyo; President: Koichi Watanabe; hereinafter referred to as “Cygames”) is pleased to announce that a joint research paper by Cygames Research, Cygames’ in-house research institute, and the Institute of Science Tokyo has been selected for presentation at the 19th European Conference on Computer Vision (ECCV 2026), a premier international research conference in the field of computer vision.

ECCV 2026 will be held from September 8 through September 12 in Malmö, Sweden.

Selected Paper

Title: Motion Style Slider: Endpoint-Supervised Continuous Style Control for Human Motion Diffusion
Authors: Chen-Chieh Liao (associate professor at the Institute of Science Tokyo, School of Computing; Cygames*), Yichen Peng (Institute of Science Tokyo, School of Computing), Yiyi Cai (The University of Tokyo), Yûi Ono (Cygames), Hiroki Hanaoka (Cygames), Erwin Wu (Institute of Science Tokyo, School of Computing), Hideki Koike (Institute of Science Tokyo, School of Computing), and Shuichi Kurabayashi (Cygames)

*Partial research conducted during an internship with Cygames between January 2024 and August 2025. Currently affiliated with the Institute of Science Tokyo.

Research Context

With the significant advances in automated editing technologies over recent years, it has become possible to emulate realistic human motion. However, conventional technologies used to adapt motion styles in an expressive manner frequently reflect them as predefined, fixed performances (such as “happy,” “sad,” “angry,” and so forth). Minute directions such as “a little more intense,” or “a little more subdued” to fine-tune the performances require iterative motion capture and compilation of data for each adjustment, which results in extensive financial and time expenditures for development teams.

* “Latent diffusion models” refers to technologies that process image or 3D motion data in a compressed space (“latent space”) to efficiently build or edit material at high quality.

Research Overview and Key Features

The research, conducted utilizing data expressly produced for the purposes of the study by the Osaka Cygames’ Motion Capture Studio, proposes an innovative technology called “Motion Style Slider” to address the aforementioned issue. Motion Style Slider is a new framework that allows for the continuous control of a motion’s performance intensity, much like a regulating slider, based on two endpoint datasets: neutral-content motion without an emotional component, and fully stylized motion. Using this technology, motions can be modulated to produce a range of subdued to exaggerated performances without the need to record intermediate performance data in advance.

Outlook

The results of the proposed research constitute a fundamental technology that enhances control over motion-style transfer frameworks in 3D animations, allowing artists and directors to more intuitively reflect their intent for the performance of a character. Moving forward, with a focus on practical applications in the production of games, video, animation, and virtual avatars, future research and development will be conducted to improve realism and reduce error in style-transferred motion, as well as to adapt the technology across various character and performance styles and to integrate it into production workflows.

About Cygames Research

Following the belief that games can be empowered with science, Cygames Research is an in-house research institute founded to take a multifaceted approach toward the creation of fundamental technologies applicable to the development of a wide range of entertainment. Collaborating with universities and research institutions both within Japan and globally, Cygames Research develops and applies theoretical frameworks in cutting-edge fields.

Website: https://research.cygames.co.jp/