For most of real-time graphics history, character hair has been a lie made of cards: flat textured strips, arranged in overlapping layers, with a hair texture and an alpha channel doing the rest. It’s cheap, it renders fast, and from a normal camera distance it works. Nearly every game character you can name has hair built this way.
Strand-based hair — where the hair is actually hundreds of thousands of individual curves — was reserved for film, because it costs far more to render and simulate. That gap has closed fast. Modern engines can render and simulate real strands, and audiences have started to notice which is which.
Which leaves a practical, unglamorous problem: a huge back catalogue of characters whose hair is cards. Converting them by hand means regrooming every character from scratch.
HairCS, from Zixuan Lu, Tongtong Wang, Yuefan Shen, Zhongtian Zheng, Chenfanfu Jiang, Yin Yang and Kui Wu (University of Utah, LIGHTSPEED and UCLA), automates that conversion.
What it does
Input: a collection of textured triangular or quad strips — an ordinary hair-card model. Output: a strand-based hairstyle that preserves the original look while adding fine geometric detail the cards never had.
The value is in the production constraints it respects, which is what separates a research demo from something a studio can use:
- Strands originate from the scalp — not floating in space, which is what naive conversion produces
- Roots are uniformly distributed across the scalp
- Hair volume is plausibly filled, so it doesn’t look like a sparse wig over a bald head
- Results work directly with strand-based rendering, physics simulation, and standard grooming modifiers — clumping, curling, noise
That last point is what makes it useful. A converted asset that can’t take a grooming modifier is a dead end; one that can becomes the starting point an artist refines.
They validate on a wide range: short and long, curly, and hard cases like buns and ponytails, which are exactly where card models carry the least real structure and conversion has the most to invent. A dataset is released alongside.
Why it matters beyond games
Remastering has become a real category of work. Studios reissue back catalogues on new hardware, and hair is one of the most visible tells of an asset’s age. An automated path from cards to strands turns a per-character regroom into a batch process with artist cleanup.
There’s also a broader pattern worth naming. A great deal of the 3D content that exists was authored under constraints that no longer apply — low polygon counts, baked lighting, painted-in detail, card hair. Upgrading that content is its own research area, and it’s an underrated one, because the alternative is remaking everything. HairCS sits alongside work like KaiNinja’s part-level generation in the same broad project: making existing assets usable under modern requirements rather than starting over.