preprint · arXiv
How to Build Digital Humans? From Priors to Photorealistic Avatars
Wojciech Zielonka, Tobias Kirschstein, Timo Bolkart, Simon Giebenhain, Vanessa Sklyarova, Xiang Deng, Donglai Xiang, Shunsuke Saito, Yebin Liu, Matthias Niessner, Justus Thies
In plain English
A state-of-the-art report covering 2021 to 2025 that maps how photorealistic 3D digital humans are actually built, splitting the pipeline into learning priors of human appearance and motion, creating a personalised avatar, and animating it — then concentrating on the first two. It walks through 3D representations (meshes, neural fields, 3D Gaussians), morphable models and generative priors, and treats head, hair, hands, garments and full body as separate problems with separate literatures. For practitioners this is the orientation document for understanding what a vendor means by "avatar": it makes clear which parts of a convincing digital human are solved, which (hair, hands, cloth) are still handled as special cases, and it closes on ethics and on open problems including efficiency, behaviour modelling and spatial audio.