This talk presents a new approach to marker-free, holistic human performance capture that eliminates the need for complex hardware, markers, or manual calibration. By leveraging machine learning models trained exclusively on synthetic data, the method enables high-fidelity reconstruction of the entire human body, including face, hands, eyes, and tongue in a wide range of environments and camera setups. The approach combines synthetic training data with parametric human models to produce stable, accurate, and world-space results without manual intervention. The talk will explore how synthetic data addresses annotation challenges, enhances generalization, and paves the way for more accessible and flexible human motion capture.