A formal descriptive language for clothing construction.
Dance has Labanotation. Music has staff notation. Architecture has plan/section/elevation conventions. GNL brings the same rigor to garments — a generative descriptive language where a valid expression is sufficient to construct a garment without ambiguity.
- Body-anchored — the body is the coordinate system, using anatomical landmarks (
@shoulder.L) and regions (%torso.front) - Topological — garments are surfaces with boundaries and openings
- Constructive — descriptions encode build order, not just final form
- Composable — complex garments are compositions of simpler elements
GARMENT t_shirt [SYM] {
FABRIC: M(160gsm, fluid, biaxial:15%, 1.0, knit.jersey)
front = P(%torso.front, contour, 1.15)
back = P(%torso.back, contour, 1.15)
sleeve = P(%arm[0..0.4], contour, 1.2)
neck = O(@neck, circle, body+8cm)
hem = O(@hip, circle, body+10cm)
BUILD:
S(front.shoulder, back.shoulder, serged)
>> S(sleeve.cap, {front.armhole, back.armhole}, serged)
>> S(front.side, back.side, serged)
>> F(hem, 2.5cm, in)
}
The language is formally defined as a PEG grammar targeting Peggy. The generated parser produces a richly-typed AST which is adapted to the renderer's internal format at runtime.
npm install # install Peggy (dev dependency only)
npm run generate # regenerate viewer/src/gnl-parser.js from grammar/gnl.peg
npm test # run parse + adapter tests against all examplesThe repo includes a live viewer that parses GNL and renders both assembled garment views and flat pattern pieces.
Write GNL on the left, see the full garment on the right — with stitch lines, dimension callouts, and construction details.
Toggle to "Pieces" to see the individual flat pattern pieces with shape outlines, grain lines, and dimensions.
The repo includes a converter that transforms garment templates from the Korosteleva NeurIPS 2021 dataset (2D panel geometry as JSON) into GNL.
# Auto-downloads 21 templates from GitHub on first run, converts all
node converter/convert.jsFour example templates (tee, skirt, pants, dress) are also available directly in the viewer — select from the "Korosteleva Dataset" section of the examples dropdown. A GNL/JSON toggle lets you compare the raw geometric input with the converted semantic output.
See converter/README.md for details on the mapping approach.
v0.2 — Draft. Includes grain parameter, directional ease, princess seams (EDGE), lining (LAYER), and component composition (USE/ATTACH). A starting point that will need refinement through use, critique, and input from garment-makers, pattern-drafters, and computational designers.
All rights reserved.