HairDeck

HairDeck is a research method for generating stylized hair for game and avatar characters from a single image. Instead of producing realistic strands or one fused mesh, it generates hair cards, the artist-style ribbons whose geometry forms the visible hair, together with the spring-bone chains that animate them.

Each card is encoded as one token covering geometry, root position and orientation. A set diffusion transformer denoises all card tokens jointly, conditioned on an image or a point cloud. A second transformer generates spring-bone chains, and geometric skinning binds the cards to them. The output carries per-card UVs and spring bones and can be exported as a VRM avatar.

According to the project page, on the test set of CHARM (the prior stylized hair-card generator) HairDeck reduces Chamfer distance by 44%, cuts card-orientation error by more than half, and generates cards 33 times faster. It also outperforms CHARM, image-to-3D pipelines and strand-based methods on single-image input. A YouTube review reports code and a roughly 2.45 GB model under Apache 2.0 that run on Nvidia and Apple Silicon; the repository link was not in the provided material. Hair colors in the demos are sampled from the input images.