Grand Challenge — Real-Time Digital Avatars — A Comparative Analysis

What does it take to make a digital human talk back to you in real time? This session examines that question through three distinct lenses: game-engine rigs like MetaHuman Creator, AI video generation, and Gaussian splatting — each representing a different philosophy for achieving responsive digital embodiment.

No single approach dominates every dimension. Some prioritize control, others realism, others speed. What’s most compelling is how these paradigms are beginning to borrow from one another — and how quickly the boundaries between them are dissolving.

Invited guest Yuntian Chai delivers a clear walkthrough of how each approach works, live demonstrations, and a thoughtful comparison of their strengths, trade-offs, and future trajectories. Hosted by Chen Zhou.

Talk — Real-Time Digital Avatars — A Comparative Analysis (Yuntian Chai)

A comparative analysis of real-time digital avatar pipelines: MetaHuman-style game-engine rigs, diffusion-based AI video generation, and Gaussian splatting. The talk covers how each works, live demos, and a comparison of control, realism, latency, and emerging hybrid approaches.

Speakers

Yuntian Chai is a multidisciplinary builder working at the intersection of art, computer science, and psychology. He is the co-founder of Cogix and has served as a founding engineer across multiple startups, leading the development of early-stage products from concept to launch.

Earlier in his career, he worked as a developer and technical engineer within the Unreal Engine ecosystem at Epic Games, contributing to developer-facing technologies. He later joined Hedra, where he prototyped the company’s real-time avatar system powered by diffusion-based video generation.

Yuntian holds degrees spanning psychology, computer science, and art, including undergraduate and master’s training in Psychology from East China Normal University, a Master’s degree in Computer Science from the University of Pennsylvania, and a Master’s degree in Interactive Media from Sheridan College. He now focuses on building AI-native creative tools that expand the boundaries of human creation and expression.

Host

Dr Chen Zhou is a Lecturer in the School of Psychology & Neuroscience at the University of Glasgow. His research combines classic psychophysical methods, cutting-edge virtual reality technologies, and computational modelling to study human perception and action in social contexts. He is particularly interested in modelling human social behaviour and translating these findings into algorithms that can be implemented in intelligent virtual agents and social robot platforms. Dr Zhou has extensive experience developing virtual reality programs using Unreal Engine and conducting behavioural experiments with robotic systems.