Real-time hair rendering leveled up.
Hair is one of the most challenging aspects in real-time graphics. A single head can easily contain 100,000 individual strands, each interacting with light and physics in complex ways. Simulating all of them is impossible in real-time without destroying performance, so for years, developers relied on shortcuts like flat textures or polygon hair cards. These solutions appeared to work at a glance but often broke down in motion, especially under dynamic lighting conditions.
With the latest wave of technology, the entire landscape changed, as developers adopted smarter, more scalable approaches. The most common is hybrid simulation, where only a few hundred or a few thousand guides are fully simulated. These guide strands form the backbone, and algorithms interpolate their motion into the remaining strands, producing the illusion of thousands of hairs at a fraction of the cost.
Now, take a look at this scene below, featuring 100 characters, each with a unique hair model of 100,000 strands, which was rasterized in just 2 ms on an NVIDIA RTX 4090 GPU (with 8× MSAA) using a novel real-time hair rendering method.
The new approach mixes hair meshes with level-of-detail techniques, and all 100 hair models together fit into only 1.7 MB of memory (13–21 KB per model). That sounds crazy, right?
The study was conducted by Cem Yuksel and his team of researchers from the University of Utah, who developed a solution for efficient GPU rendering of strand-based hair through on-the-fly geometry generation, achieving orders-of-magnitude reductions in storage and memory bandwidth.
Cem Yuksel
The team used mesh shaders to distribute computation effectively and employ a custom texture layout to offload parts of the work to hardware texture units. They also introduced procedural styling operations to generate strand variations across diverse hairstyles and a consistent coordinate-frame system to attach these variations to animated or deforming meshes.
Cem Yuksel
The team claims to have achieved unprecedented performance for strand-based hair rendering, with hundreds of full hair models animated and displayed in real-time on a consumer-grade GPU.
If that still sounds too complex, check out a quick video from Two Minute Papers explaining the thing:
You can find the research materials here. Also, join our 80 Level Talent platform and our Discord server, follow us on Instagram, Twitter, LinkedIn, Telegram, TikTok, and Threads, where we share breakdowns, the latest news, awesome artworks, and more.