HairDeck: Animatable Hair from a Single Portrait

Researchers from Tsinghua University, Carnegie Mellon University, and CUHK introduced HairDeck (arXiv:2610.40599) — a method for generating stylized hair cards and spring-bone chains for hair animation from a single image, which is more accurate and approximately 33 times faster than the autoregressive CHARM generator on the CHARM test set.

Compiled by Sergey KostenchukPublished 2026-11-01Updated 2026-11-01

2026-11-01 Multimodal NVIDIA

🎨 HairDeck: Animatable Hair from a Single Portrait

A team from Tsinghua, Carnegie Mellon, and CUHK presented HairDeck (arXiv:2610.40599): each hair "card" is encoded by a single token, a set diffusion transformer creates the entire set at once, and a second model constructs spring-bone chains to which the cards are attached via skinning. On CHARM tests, it is more accurate (Chamfer -44%): ~1.01 s vs. 33.7 s — approximately 33 times faster.

🌍 The method addresses a bottleneck in creating stylized 3D characters: the only previous card generator, CHARM, operated autoregressively. Parallel generation of the entire set accelerates production pipelines for VTuber avatars, games, and virtual characters.

👤 Demo on Hugging Face Space: from a single portrait — ready-to-use hair with physics and VRM export in seconds. Code and weights are open under Apache-2.0. Verified so far only on hair cards.

Source 1: https://arxiv.org/abs/2610.40599 Source 2: https://hyzcluster.github.io/HairDeck

Sources