The Slug text patent went public domain in March. Here's what to do

A patent that stopped being a constraint

On March 17, 2026, Eric Lengyel dedicated his patent on the Slug text rendering algorithm to the public domain. He had published the algorithm in 2017 and patented it in 2019. The method draws glyphs directly from their outlines on the GPU, with no texture atlas and no per-frame tessellation. According to Alpha Pixel, a company that has built a C++20 implementation called Slughorn, that change is what allows them to discuss and ship the technique openly. If your product draws text in 3D, or has been avoiding the algorithm for licensing reasons, the legal question looks different now. The engineering question, which renderer fits your product, has not changed.

Why glyphs are hard to draw on a GPU

A font does not store letters as pictures. It stores outlines: closed loops of straight lines and Bezier curves, which are smooth curves defined by a few control points. A pixel is inside the letter if a ray shot from it crosses the outline in a way that gives a nonzero winding number, a tally of how the outline wraps around that point.

The GPU has to fill the interior, antialias the edges, and stay sharp at 8 pixels in a menu or at full screen on a billboard seen in perspective, all while drawing thousands of glyphs per frame. Most engines avoid the hard version by baking glyphs into textures ahead of time. Each baking method makes a different tradeoff:

• Texture atlas: Rasterize each glyph once into a shared texture. Fast and portable, but it blurs when enlarged, shimmers when shrunk, and needs a new atlas for every size. Chinese, Japanese and Korean fonts, with tens of thousands of glyphs, make this a memory problem.
• SDF (signed distance field): Each texel stores the distance to the nearest edge, so a small texture scales up cleanly. The catch is that interpolation rounds off sharp corners.
• MSDF (multi-channel SDF): Three channels encode distances to different edge subsets, and the shader takes the median to keep corners crisp. Better quality, but still a baked atlas with generation cost and fixed resolution.
• Tessellation and coverage (Loop-Blinn, NV_path_rendering, Pathfinder, Rive): Turn outlines into geometry the GPU rasterizes. Resolution independent, but geometry must be rebuilt when it changes, and antialiasing or hardware dependence can be difficult.

What Slug does differently

Slug keeps each glyph as a short list of curves in a small GPU buffer, plus a per-glyph structure that splits the glyph into horizontal bands so a pixel only checks nearby curves. In the fragment shader, it casts a ray per pixel, finds where it crosses those curves, and counts crossings to get coverage. Lengyel's key contribution, per the source, is a precise rule called root eligibility that decides which curve-ray intersections count, so the math stays exact where curves join and avoids cracks or double-counts.

Because coverage is solved analytically after the transform, nothing is baked. The source's claims follow from that: the same glyph stays sharp at 6 or 6000 pixels, survives perspective, handles huge character sets without an atlas, and allows text to change every frame, all in a single draw with an ordinary shader. These are the vendor's claims about its own approach, supported by its own side-by-side renders. In those, the texture-based panels got 64 texels per em. Slughorn and the distance-field methods held up under grazing perspective, the bitmap blurred, and Rive's renderer, which accepts only 2D affine transforms, distorted the shape. At extreme zoom, only Slughorn and Rive stayed clean.

Questions You Should Be Asking

• Can our users actually leave the size and angle range we baked for? If not, MSDF may already be good enough, and switching is cost with no benefit.
• Where is the benchmark? The source shows image comparisons of one capital R, not frame times, memory use, or shader cost on our target hardware. What does per-pixel curve solving cost on our lowest-end device?
• Who is selling us this? The comparison comes from the maker of a Slug implementation. What would an independent test show?
• Does the patent dedication cover our situation? It addresses the algorithm's patent. It does not settle other licensing, such as font licenses or an implementation's own terms. Has counsel confirmed?
• What does our text pipeline actually look like? If strings are static and sizes bounded, why rebuild a working atlas system?

What To Watch Next

The signal is independent measurement: published frame times and memory figures for Slug-style rendering against MSDF on mobile and VR hardware, and whether engines beyond Alpha Pixel's osgSlug integration adopt it now that the patent is public domain. Until that data exists, a sensible operator prototypes it only where text meets an uncontrolled camera, in 3D, AR, VR or CAD views, and leaves working atlas pipelines alone.