TAC-Twin: A Rapid Framework for Personalized Doppelgänger Avatar Creation Using a Modular Virtual Human Pipeline

We present an end-to-end framework for rapidly creating interactive, personalized avatars for scalable training and simulation applications. Built as an extension of the Virtual Human Toolkit, the framework integrates technologies for audio-visual sensing, speech recognition, natural language processing, nonverbal behavior generation, and high-fidelity text-to-speech synthesis. A personalized avatar is defined here as a real-time, embodied digital representation of an actual individual rather than a generic character. The creation pipeline requires only a single facial photograph, processed through a photorealistic character generation workflow, then refined, customized, and deployed in a real-time 3D environment for integration with conversational AI and synthetic voice generation. The system also supports rapid generation of generic avatars from high-quality synthetic headshots produced by generative AI, enabling the creation of diverse, realistic or stylized cohorts within minutes. Our initial use case examines whether personalized avatars enhance engagement, motivation, and performance compared to generic avatars, with the hypothesis that personalization increases relevance, identification, and learning outcomes. We describe the architecture, avatar creation pipeline, and role of generative AI in accelerating development, and share early implementation insights.