Interview with Yale computer science professor Theodore Kim, leading landmark research in hair animation funded by The Bungie Foundation.
What inspired this research, and what problem were you trying to solve when it comes to representing African-American hair in videogames?
We started this research after the 2020 murder of George Floyd, as all the racial injustices during the pandemic came to the fore. In collaboration with Professor A.M. Darke at UCSC, we saw that CGI algorithms for Afro-textured hair were entirely missing. As a result, what little Afro-textured hair that did appear was often not respectful, and anything that was respectful was very difficult to make. The digital tools were all geared to help artists create and style straight hair.
Why has afro-textured hair historically been difficult to model and animate accurately in games and animation?
The default assumption for decades has been that straight hair is the most important type of hair, and all other types are just minor variations. As a result, decades of research has built digital tools for straight hair, whose performance gets worse and worse for curlier hair. When computer graphics was getting started in the 1980s, everything was hard. Both straight and curly hair were equally hard. Nowadays, straight hair is much easier since lots of people have developed algorithms for it. Afro-textured hair in many ways has been stuck this whole time at the 1980s starting line.
What makes this research approach different from how hair has traditionally been handled in computer graphics?
We have discarded the assumption that hair is just a straight rod, so we need to rewind and ask: "What if it's not a rod? What if it's a helix?" The basic math that gets pulled in to examine these questions then changes.
Your paper introduces new ways to describe the shape and movement of coily hair—how would you explain that work in simple terms to a general audience?
We have found three different shape classes that are unique to coily hair. First, very coily hair forms a spongy layer near the scalp, but then coalesces into chunky coils further from the scalp. We have devised a way to model the spongy layer in a way that looks authentic, as well as a way to transition to coils. We call this "phase locking". Second, each of these coils contain a bunch of stray hairs that occasionally leap in and out of the helix. Depending on how often they leap out, the hair can look dull and frizzy, or slick and shiny. We came up with a way to model these different looks called "period skipping". Third, coily hair can contain staple-shaped curves at moments when the coil changes direction. Anybody who ever fussed with an old telephone cord recognizes this shape. We came up with a way of computing it, which we call "switchbacks".
How does improving the underlying technology change what artists and developers are able to create?
Some of these shapes, such as the switchbacks, were simply impossible to compute before. Nobody, even the top Afro-textured hair artists in the world, knew how to get these shapes into their models. They knew it was missing, but just had to go without. Now with the help of our new algorithms, they can finally get the look they want.
Why is it important that representation challenges like this are solved at the tool and technology level, not just through extra manual work by artists?
There are basic questions of algorithmic equity that have been neglected for decades. To draw a painting analogy, it's like artists who wanted to paint pictures of straight hair had decades of scientists designing custom paintbrushes, air brushes, varnishes, and canvas materials just for them. In contrast, if you wanted to paint Afro-textured hair, all you got was two paintbrush types, both of which were originally designed for straight hair. You can make it work with lots of time and sweat, but there's still the basic question of fairness.
How does this research connect to broader conversations around inclusion and equity in gaming and digital spaces?
One of the reasons that Afro-textured hair algorithms have not been developed for so long is that Black scientists and artists have been historically excluded from the film and gaming spaces. There were no voices well-positioned to point out these glaring algorithmic deficiencies.
In the film and gaming space, it goes back to the basic question of who gets to be viewed as "human", and which version of humanity is worthy of algorithmic inquiry. If you do not qualify for such inquiries, secondhand algorithms get used to generate a lesser version of your likeness on a screen.
Looking ahead, how could this work shape the future of character creation and player representation in videogames?
We hope that when people look back years from now, it will be painfully obvious how narrow and limited today's character creation tools were. We hope that all the current looks and styles of Afro-textured hair become so easy to make that future developers can push the envelope and inspire all sorts of gorgeous new designs.
What's next for your and your team?
There's lots of work still to be done. Basic questions on how best to generate, style, and animate hair, and make it fast enough that some of our improvements can start appearing in games, still need to be answered.
Why was support from the Bungie Foundation important in making this research possible, and what role can philanthropy play in advancing more inclusive technology in gaming?
None of this would have happened without the support of the Bungie Foundation. The other main source of funding for this sort of research is the federal government, but that has been dramatically shrinking over the last year. For scientists who still believe that inclusion is an important ingredient for a better world, partnering with philanthropy is the only way we can keep pushing forward.
