On June 12, 2025, VinUniversity hosted a seminar titled “StyleGAN-Based 3D Stylized Face Generator & LLM-Based XR Applications” in Room I201. The event featured Professor Cha Sungdeok from VinUniversity, Professor Seungyong Lee from POSTECH (Korea), and Dr. Chang-Hun Kim, CEO of DeepXRLab. It attracted a highly engaged audience of students, researchers, and faculty members.

Seminar Highlights

Expert Presentation 1 – Prof. Seungyong Lee (POSTECH, Korea):
Professor. Seungyong Lee introduced Toonify3D — a cutting-edge framework for generating stylized 3D cartoon faces from 2D images using StyleGAN. A key innovation of the research is its ability to perform “3D lifting” from 2D stylized faces, despite the absence of standard datasets. The team proposed a novel method for establishing semantic correspondences between stylized and realistic faces, enabling the construction of synthetic datasets and the training of a neural network called StyleNormal for predicting surface normals. The system exhibits strong cross-domain generalization and can generate full-head 3D models with remarkable accuracy. Moreover, the framework supports GAN-based facial expression editing, offering broad creative potential. Several techniques have already been successfully transferred to Adobe, underscoring the project’s industrial impact.

Expert Presentation 2 – Dr. Chang-Hun Kim (DeepXRLab):
Dr. Chang – Hun Kim presented an innovative solution that integrates Large Language Models (LLMs) into Extended Reality (XR) platforms. Developed by the DXR Lab team, the system leverages specialized lightweight LLMs (sLLMs) to power intelligent, interactive AI avatars with real-time voice, emotion, pose, and facial synchronization. The avatars are highly customizable in conversation style, personality, and personal data integration. Using advanced techniques such as emotion transfer and transformer-based facial animation, these avatars deliver engaging and lifelike interactions. The applications span multiple domains—from AI HealthMate, a virtual healthcare consultant, to AI-powered K-pop stars and gamified emotional chatbots. The solution also demonstrates scalability and international viability, with successful expansion into the Japanese market. With its solid foundation in academic research and product development, DXR Lab is at the forefront of generative AI and immersive content innovation.

Acknowledgements

We extend our sincere thanks to Professor Cha Sungdeok for his coordination and support in organizing this seminar. Our heartfelt appreciation also goes to Professor Seungyong Lee and Dr. Chang-Hun Kim for their thought-provoking and visionary presentations. From Toonify3D, which introduced a novel method of stylized 3D face generation, to the integration of LLMs into XR platforms that reimagine human–AI interaction, both talks embodied the spirit of innovation and interdisciplinary collaboration that VinUniversity champions. Finally, we thank all participating researchers and students for their enthusiastic engagement and forward-looking perspectives.

SOME MEMORABLE MOMENTS FROM THE EVENT