Living Hinge Generator — Operating Guide

Parametric lattice-hinge ("living hinge") SVG generator for laser cutting. Two files, which must sit in the same directory:

Requires Node ≥ 10.12. No dependencies, no install step.

The stock is 3mm Baltic birch plywood — that is what these are cut in, and what the defaults are aimed at. The tuning numbers are still starting points, not tested values: they are derived from geometry rather than measured off cut material, and bridge width in particular decides whether a hinge flexes or snaps. Cut a coupon (see Finding your real numbers) before committing a full panel.

Get the files

• Everything as a ZIP — the CLI, the library, this guide, all eight cut-ready examples and the test coupons.
• Repository — the code, if you want to change a pattern or read how it works.

Released under CC0 1.0 — do what you like with them, no attribution needed. Built for LaserMadeMusic.

1. What it actually does

It emits an SVG containing a rectangular panel outline plus a field of cuts that let a rigid sheet bend along one axis (or two, for the biaxial patterns). It does not do nesting, tabs, joinery, or kerf compensation — it produces the hinge field and the panel rectangle, and you compose that into a larger part.

Everything is computed in millimetres internally. --units in only changes what gets written out.

Panel convention

The slits are real generator output, not drawn by hand — see panel-convention.js. Proportions are illustrative; the shipped examples are 50 × 200mm.

• --width — the span the slits run across. The bend axis is parallel to this.
• --length — the direction the panel rolls up. More length = more rows = gentler curve.
• --margin — solid end tab at each end of the length. This is your glue/fixing surface.

The rule everything else serves: slits must reach x=0 and x=width. An uncut rail down either long edge stiffens the entire panel and it will not flex. The generator enforces this — it is why --side and --cell get resized (§6) and why the crosshatch edge cuts are exempt from --minSliver.

2. Quick start

cd living-hinge          # wherever you cloned or unzipped it

./living-hinge.js                                  # 50 × 200mm straight, to ./
./living-hinge.js -p wave -w 80 -l 300             # wave pattern, 80 × 300mm
./living-hinge.js -p dogbone -w 2in -l 10in        # inch input, mm output
./living-hinge.js -p chevron -w 3in --units in     # inch input AND inch output
./living-hinge.js --all --outdir ./hinges          # all 8 patterns at one size
./living-hinge.js --list                           # pattern names + their defaults
./living-hinge.js --help                           # full option list

Add --dry-run to any command to see the stats and the filename it would write without touching the disk. Use it constantly — it is instant and it is how you iterate.

./living-hinge.js -p dogbone --bridge 2.5 --hole 0.4 --dry-run
# dogbone  slit=10.5  bridge=2.5  hole=0.4  holeDia=0.8  ligament=1.7  pitch=3  rows=62 ...

As a library

var gen = require('./living-hinge-generator');
var r = gen.generate({ pattern: 'wave', width: 80, height: 300, bridge: 1.5 });
// r.svg     -> the SVG string
// r.stats   -> { slit, bridge, wavelength, ..., pattern, cuts }
// r.options -> every resolved option, after defaults and fitting

Note the library says height, the CLI says --length. Same thing.

3. Complete option reference

Panel

Tuning (pattern-dependent)

An option that does not apply to your pattern is reported on stderr and ignored, not silently dropped:

living-hinge: --hole does not apply to straight; ignoring it

Output & safety

--maxCuts / --maxPoints exist so a mistyped feature size fails in a second instead of exhausting memory. You will only ever touch them for a genuinely enormous panel.

Input formats

• Bare numbers are millimetres: --width 80
• Suffixes convert that one value: --width 2in, --width 3cm, --width 80mm, --width 2"
• A suffix does not change output units. Only --units in does that.
• --segments, --maxCuts, --maxPoints are counts: no suffix, no fractions.
• --flag=value and -w80 and -w 80 all work.

flat-part-check does not fit a hinge, and that is not a fault

Running the flat gate in lasermade-tools over this folder gives 16 files, 115 checks, 21 failed. Every one of those failures is the tool's model, not the files:

• cut paths are closed counts all 279 paths in a panel as open, and they are. A slit is a single cut LINE: it frees nothing and has no inside. The check is for an outline that must meet itself, which a hinge has none of.
• holes are inside the outline and hole is big enough assume one part per sheet with holes in it. A coupon sheet carries several.
• millimetre-true, ink is in the palette and black frees the part fail on panel-convention.svg alone, which is the documentation diagram and not a cut file at all. --dir sweeps it up.

What the tool IS worth running for here is fits the bed, and for that pass --min-edge 1.5, since its 3mm default is for a bullroarer's cord hole under string tension and nothing here is.

Every number here is a DRAWN number

Nothing in this generator compensates for the kerf, and neither does anything in the trumpet's bell or mouthpiece; only the two bore generators do. For a hinge that matters more than for most parts, because the dimension carrying the load is the bridge, and a bridge sits between two slits which each open by half a kerf as the beam goes down them.

On a laser measured at 0.13mm on 2026-09-09, the shipped sweep reads:

drawn 1.00  ->  0.87mm of ply actually left
drawn 1.25  ->  1.12mm
drawn 1.50  ->  1.37mm
drawn 1.75  ->  1.62mm
drawn 2.50  ->  2.37mm

This does not invalidate the sweep. Its whole method is to cut the coupons, flex them and keep the one that survives, and that answer is right whatever the label says. It matters when the number travels: ask for --bridge 1.5 in a later design expecting 1.5mm of ply and you get 1.37, and on a different laser you get something else again. Record the coupon that worked by its DRAWN value and the kerf you cut it at, and the pair transfers; the drawn value alone does not.

4. Suggested starting numbers

Unvalidated. Treat as a first guess to bracket with a coupon test.

The two knobs that matter most:

• --bridge is the ligament. It is the uncut material between collinear slits, and it is what actually bends and what actually breaks. Too narrow → snaps. Too wide → the panel stays rigid. This is the number to sweep first.
• --pitch is the row spacing. Smaller pitch = more rows over the same length = tighter achievable radius and a smoother curve, at the cost of more cutting time and a weaker panel overall.

--targetSlit sets slit length indirectly: the generator picks segments = round(width / targetSlit), so slit length stays near your target as panels get wider. Prefer it over --segments, which hardcodes a count that stops making sense at a different width. If you pass both, --segments wins and you get a warning.

5. The eight patterns

Each is committed cut-ready in examples/ at 50 × 200mm — the size every figure below is quoted at — so the numbers in each section describe a file you can open. All eight span the full panel width, x = 0 to x = 50, which is the rule §1 says everything else serves.

previews/ holds a display rendering of each — gold is the panel that stays, cream the slits the laser removes — with the cut stroke thickened and painted onto the panel. Browse those; cut the ones in examples/. A real cut file is a 0.02mm hairline on no background at all, which most viewers show almost invisibly against a transparency checkerboard.

straight	dogbone	wave	chevron
torsional	honeycomb	auxetic	crosshatch

Click any one to download its cut file.

Stats below are real output at the default 50 × 200mm, margin 6.

straight — the default; start here

Plain staggered slits. Fewest cuts, fastest to cut, strongest for a given bridge.

====  ====  ====  ====        row A
  ====  ====  ====  ====      row B (offset half a period)
====  ====  ====  ====        row A

slit=11 bridge=2 pitch=3 rows=62 segments=4 → 279 cuts, 9.0 KB

• Use when: single-axis bend, and you have no reason to use anything else.
• Optimal-ish: --bridge 1.5 --pitch 3 for 3mm ply. Drop pitch to 2.5 for a tighter curve.
• Stress concentrates at the slit ends. If it cracks there, switch to dogbone.

dogbone — straight, with the stress risers removed

Adds a relief hole at each slit end. This is the fix for cracking at slit tips.

slit=10.25 bridge=3 hole=0.5 holeDia=1 ligament=2 pitch=3 rows=62 → 713 cuts, 28.5 KB

• --hole is a RADIUS. The stats report both hole (radius, matching the flag) and holeDia.
• Watch ligament — the stat that predicts failure. ligament = bridge − 2 × hole, and it is the real load-bearing width. The generator refuses ligament ≤ 0.
• Because the holes eat into it, dogbone needs a wider --bridge than straight for the same strength. Default is 3 vs straight's 2 for exactly this reason.
• Optimal-ish: --bridge 3 --hole 0.4 → ligament 2.2. Raise the hole only if it is still cracking at the tips.
• ~2.5× the cuts of straight. Noticeably longer cut time.

wave — sinusoidal slits

Longer ligament path for the same pitch, so it bends more easily than straight, and the curved path distributes stress instead of concentrating it.

slit=11 bridge=2 wavelength=13 amplitude=1 pitch=3 rows=62 → 279 cuts, 74.3 KB

• All rows share the sine phase, so curves stay parallel and the spacing never closes up.
• Largest file of the eight (74 KB) — each slit is a polyline. --step controls resolution: 0.5 default, 0.25 for a visibly smoother curve at ~2× the file size. Below ~0.1 you are cutting finer than the beam and just making the file huge.
• Optimal-ish: --wavelength 13 --amplitude 1 --bridge 1.5. Keep amplitude well under half the pitch or adjacent rows crowd each other visually (they never actually collide — shared phase guarantees it).

chevron — zigzag slits

Wave's behaviour with straight-line segments. Cuts faster, file is 4× smaller (17 KB vs 74 KB), sharper look.

slit=11 bridge=2 wavelength=13 amplitude=1 pitch=3 rows=62 → 279 cuts, 17.2 KB

• Use when: you want wave's compliance but not its file size or cut time.
• Optimal-ish: same as wave. This is the better default of the two unless you specifically want the curve.

torsional — I-cuts (slit with perpendicular end caps)

Each slit gets a short perpendicular cap at each end, lengthening the torsion path. More compliant than straight at the same bridge.

slit=11 bridge=2 cap=1 pitch=3.5 rows=53 → 608 cuts, 20.3 KB

• This is an I-cut, not a spiral/S-flexure. Spirals do not tile cleanly to the panel edge.
• Hard constraint: 2 × cap ≤ pitch. Beyond that, caps from adjacent rows meet and the panel is sliced into strips. The generator refuses.
• Default pitch is 3.5 (not 3) to leave cap room.
• Optimal-ish: --cap 1 --pitch 3.5. To go more compliant, raise both together — --cap 1.5 --pitch 4.

honeycomb — hex lattice (biaxial)

Compound curvature: bends on both axes. Hex vertices stay uncut and act as the hinge nodes.

side=4.8113 gap=1.2 slit=3.6113 → 481 cuts, 23.6 KB

• --side is a target. Delivered 4.8113 from a requested 5 — see §6.
• --gap is the uncut node. It must be shorter than the fitted side or there is no cut left. Smaller gap = more compliant and more fragile.
• Optimal-ish: --side 5 --gap 1.2. For more flex, --gap 0.9; expect the nodes to become the failure point.

auxetic — rotating squares (biaxial)

Squares joined only at their corners; those corners rotate. Negative Poisson's ratio — it expands laterally when stretched, so it forms domes rather than cylinders.

cell=6.25 gap=1.5 slit=4.75 → 518 cuts, 20.0 KB

• Use when: you need a dome/saddle, not a roll.
• --cell is a target (fitted, §6). At 50mm wide, 6.25 divides evenly so you get exactly what you asked for.
• Optimal-ish: --cell 6.25 --gap 1.5. Smaller cells = finer, more compliant, more cuts.

crosshatch — alternating short slits (biaxial)

Checkerboard of horizontal and vertical slits. Fewest cuts of the three biaxial patterns and the simplest to reason about.

cell=6.25 slit=4.5 cells=8 x 30 edgeSlits=30 → 240 cuts, 11.4 KB

• Outer-column slits are deliberately run out to the panel edge. Those specific cuts are exempt from --minSliver — dropping them would restore the edge rail.
• Hard constraint: slit < 2 × cell. At or beyond that, neighbouring slits merge into one continuous cut and the panel becomes a comb. Refused.
• Optimal-ish: --cell 6.25 --slit 4.5 (slit ≈ 0.72 × cell). Raising slit toward the cell size increases compliance; past the cell size you are into the merge risk zone.

Picking one

6. --side and --cell are targets, not guarantees

The three lattice patterns resize their cell so the tiling divides the panel width exactly, putting a cut on both long edges. Without it you get an uncut rail up to ~1.2mm wide down one side, and the panel will not flex.

width 30mm   --side 5 → 4.9487   (1.0% off)     --cell 6.25 → 6.00 (4.0% off)
width 50mm   --side 5 → 4.8113   (3.8% off)     --cell 6.25 → 6.25 (exact)
width 75mm   --side 5 → 5.0943   (1.9% off)     --cell 6.25 → 6.25 (exact)
width 100mm  --side 5 → 5.0204   (0.4% off)     --cell 6.25 → 6.25 (exact)

Formulas: auxetic/crosshatch use cell = width / round(width / cell); honeycomb uses side = 2 × width / (√3 × m).

Deviation is typically under 4%. The delivered value appears in the stats line, the SVG <desc>, and the filename — so a file is always named for the geometry it contains. If you need an exact cell size, pick a width that is a whole multiple of it.

7. End tabs — and what happens when you minimise them

--margin is the solid band at each end of the length axis. It is what you glue, screw, or slot into the rest of the assembly. Cuts never enter it.

You never get a zero end tab, even at --margin 0. Rows are centred in the band with a half-pitch offset, so the first row lands at least pitch/2 from the edge. At --margin 0 with pitch 3 the first row sits at y=2.5, leaving a 2.5mm solid strip. The effective tab is:

effective tab  =  margin  +  (leftover ÷ 2)  +  pitch/2      where leftover = span − rows × pitch

So the practical floor for the row-based patterns is about pitch/2 — roughly 1.5mm at default pitch. Two of the three biaxial patterns behave differently: their cuts are clipped straight at the band edge, so at --margin 0 honeycomb and auxetic really do cut to y=0 and the panel has no tab at all. Crosshatch is not one of them — its slits sit inside their cells rather than being clipped, so it floors at (cell − slit) ÷ 2: 0.875mm at the defaults, 1mm at --cell 10 --slit 8. That is under the pitch/2 floor of the row patterns and well under the 3mm glue surface recommended below, so treat it as no tab even though it is not zero.

Guidance

• Don't go below ~3mm if the tab is a glue surface. A 2.5mm strip in 3mm ply, with a full-width slit immediately inboard of it, is a tear-out waiting to happen.
• 5–8mm is a sane general range; 6 is the default for that reason.
• Go larger (10–20mm) if you are screwing, bolting, or slotting through the tab, or if the hinge carries load in tension.
• --margin 0 is legitimate when the hinge field runs into a larger part that provides its own fixing, or when you will trim the ends after cutting. Just know you are giving up the attachment surface.
• Shrinking the margin buys you rows — 6→0 gains 4 rows on a 200mm panel, about 6%. That is rarely worth losing the tab for. If you need more rows, lengthen the panel or reduce the pitch; both are much better levers than eating the tab.

8. Sizing the hinge for a bend radius

A lattice hinge approximates an arc, so the hinge field length is arc length:

hinge length  =  radius × angle(radians)
rows          =  hinge length ÷ pitch
panel length  =  hinge length + 2 × margin

Read the other way — what a given panel can do at pitch 3, margin 6:

A panel can bend tighter than the figure in the last column — you are then asking fewer rows to take more angle each, which is exactly what breaks ligaments. Treat that column as "the radius this panel reaches comfortably", and if you need tighter, add rows (longer panel or smaller pitch) rather than forcing it.

Fewer than ~6 rows in the bend behaves like a crease, not a curve.

9. Kerf

The generator emits zero-width centrelines and does no kerf compensation. Your beam removes material on both sides of every line, so the real ligament comes out narrower than the nominal --bridge.

actual ligament  ≈  bridge − kerf                       (straight, wave, chevron, torsional)
actual ligament  ≈  bridge − 2 × hole − kerf            (dogbone)

Typical CO₂ kerf in 3mm ply is 0.15–0.20mm. Dogbone at defaults:

(bridge 3, ligament in mm)

Add your kerf to --bridge if you want the nominal ligament you designed. Measure kerf once on your machine and stock: cut a 20mm square, measure it, the shortfall is one kerf.

10. Finding your real numbers

The defaults are geometry, not experience. One afternoon of coupons replaces all the guessing in §4.

cd living-hinge          # wherever you cloned or unzipped it
mkdir -p coupons
for b in 1.0 1.25 1.5 1.75 2.0 2.5; do
  ./living-hinge.js -w 40 -l 120 --bridge $b --outdir ./coupons
done

Each lands in its own file, since differing tuning goes into the filename:

living-hinge-straight-40x120mm-bridge1.svg
living-hinge-straight-40x120mm-bridge1.25.svg
living-hinge-straight-40x120mm-bridge1.5.svg
living-hinge-straight-40x120mm-bridge1.75.svg
living-hinge-straight-40x120mm.svg            <- bridge 2.0, the default, keeps the plain name
living-hinge-straight-40x120mm-bridge2.5.svg

These six are committed, in coupons/, along with a single 290 × 132mm sheet holding all of them — coupons/coupon-sheet-bridge-sweep-290x132mm.svg — and coupons/README.md, which records the settings held constant across the sweep. You can cut the sheet without running anything.

Cut all six on one sheet, in the grain orientation you will actually use. Then:

1. Bend each to 90°. Note which snap immediately.
2. Cycle the survivors 20–30 times. Fatigue is what kills living hinges in service, not the first bend.
3. Bend one to destruction to find the margin you actually have.
4. Record the narrowest bridge that survives cycling, then add 0.25mm as your working value.

Repeat per material and per thickness. Grain direction matters in ply — slits running across the grain behave differently from slits running along it, so test the orientation you will build in.

When you have real numbers, say so and they will be written into DEFAULTS in the generator and into the project memory, so future sessions start from measurements instead of estimates.

11. Output format

<?xml version="1.0" encoding="UTF-8"?>
<svg xmlns="http://www.w3.org/2000/svg" version="1.1"
     width="50mm" height="200mm" viewBox="0 0 50 200">
  <title>Living hinge - straight - 50mm x 200mm</title>
  <desc>1 user unit = 1mm. Panel 50mm x 200mm, solid end tabs 6mm. Pattern: straight; ...</desc>
  <g fill="none" stroke="#000000" stroke-width="0.02" stroke-linecap="butt">
    <g id="outline"><rect x="0" y="0" width="50" height="200"/></g>
    <g id="hinge-slits"> ... </g>
  </g>
</svg>

• Physical size is declared (width="50mm"), with viewBox at 1 user unit = 1mm. Import at 100% and it is correct size.
• Two groups. #outline is the panel rectangle — delete it when embedding the hinge into a larger part. #hinge-slits is the cut field.
• fill="none", black hairline 0.02mm stroke, so cutter software reads vector cut rather than raster engrave.
• With --units in: width="1.9685in", viewBox in 1/96 in units, stroke 0.0756 (still 0.02mm). Geometry is identical — only the units change.
• Coordinates are rounded to 4 decimal places.
• The <desc> carries the full parameter set, so a stray SVG is self-documenting. Open it in a text editor to recover what generated it.

Filenames

living-hinge-<pattern>-<width>x<length><units>[-<tuning>].svg

Default runs stay short (living-hinge-straight-50x200mm.svg). Any option differing from that pattern's default is appended (-bridge3, -margin20, -pitch2.5-bridge4), so iterating never overwrites the previous variant. Fitted values are recorded as delivered (-side5.7735, not -side6).

Existing files are never overwritten without --force.

12. When it refuses

The generator declines anything that would destroy the panel rather than cutting it. Every message names both the offending option and the way out.

Set LIVING_HINGE_DEBUG=1 for a stack trace on unexpected errors.

13. Scripting

• Exit 0 on success, 1 on any failure.
• With --all, an unbuildable pattern is reported on stderr and skipped; the rest still generate and the exit status is 1. Check it — a partial run looks like a success in the file listing.
• Stats go to stdout, warnings and errors to stderr.

./living-hinge.js --all --width 5 --outdir ./tiny || echo "some patterns failed"

14. Gotchas

• CLI --length = library height. Easy to trip over when moving between the two.
• --hole is a radius. The stats report hole (radius) and holeDia (diameter).
• --segments silently outranks --targetSlit — you get a warning, but segments wins.
• A unit suffix on an input does not change the output units. --width 2in gives a 50.8mm panel written in mm. You need --units in for inch output.
• --out will not accept a directory; use --outdir.
• --all refuses --pattern rather than guessing which you meant — it does not quietly ignore one of them.
• Wave files are large (74 KB at defaults, ~4× chevron). Some cutter software is slow with thousands of polyline points.
• The outline rect is not counted in cuts — it is the panel boundary, not a hinge cut.
• Defaults are aimed at ~3mm ply/acrylic. At a very different thickness, start from §4 rather than the shipped defaults.

15. Known limitations

• No kerf compensation. Centrelines only (§9).
• No nesting, tabs, or joinery. Hinge field and panel rectangle only.
• Bend radius is not enforced. Nothing stops you generating a hinge too short for the curve you want; §8 is guidance, not a guard.
• Material behaviour is not modelled. The generator knows geometry, not grain, moisture, ply quality, or acrylic brittleness.
• Biaxial patterns have uncrossed bands — vertical lines through the node columns that cross no cut. These are the rotating hinge nodes and are inherent to those lattices, not a defect.
• The defaults remain untested against cut material. This is the single biggest gap, and only a coupon closes it.