A fabric library that carries its own measurement record

Laboratory measurement of cloth is mature. Entering those measurements into a garment simulator is not — and the gap is where simulated garments quietly stop describing the fabric a designer actually holds.

For each CLO3D fabric input, this workspace records whether the value could be entered directly from a physical measurement, through a proxy, through calibration against a physical reference, or not at all — and what that difference does to a simulated garment. The record travels with every fabric in the library, and it is exported with every row of data.

The pipeline is four steps. Physical fabric, measured to ISO procedures (mass ISO 3801, thickness ISO 5084, tensile ISO 13934-1, n = 5). Calibration into CLO3D — bending to a 41.5° cantilever chord, stretch from a 5% secant. A measured .zfab that loads in CLO3D. Then the same garment simulated twice, once on a named library preset and once on the measured file, and compared vertex by vertex.

The dataset is seven specimens measured and digitised inside one MA project. The workspace is presented as the private material library of a fictional brand product-development team; v1 demonstrates one intake channel end to end, and the remaining channels are designed concepts, labelled as such on the page.

Three of five engine inputs have an ISO source

A standards-compliant test report cannot simply be typed into a garment simulator. Some inputs have a standard behind them; some are reached through a non-standard procedure; one is left wherever the engine put it.

• Mass per unit area — ISO 3801, enters CLO3D directly.
• Thickness — ISO 5084, enters directly in millimetres.
• Low-load stretch — derived from the ISO 13934-1 tensile curve at a 5% secant, then mapped onto CLO3D's dimensionless stretch index. A proxy, not a standard stretch test.
• Bending — calibrated against a 41.5° cantilever chord. Non-standard: this is a procedure built for the purpose, not an ISO bending test.
• Shear — left at the CLO3D default. No measurement enters.

One knit moves the dataset on its own

Across seven fabrics the average preset-to-measured difference is 24.5 mm on a drape-dominant garment — but the median is 12.0 mm, because a single cotton jersey sits at 95 mm and the other six fall between 9.9 and 21.8.

The design rule that falls out of it: measure bending when the garment is drape-dominant; measure stretch for any knit or elastic fabric, whatever the silhouette. Stretch differences survive fitting — the jersey still differs by 48.8 mm on the fitted H-line, 51% of its drape figure. Bending differences largely do not: five bending-driven fabrics fall to 16–36%, and their fitted deltas of 2.2–4.3 mm sit inside the roughly 10 mm per-vertex solver repeatability band, so those ratios are indicative only. The chiffon ignores the pattern entirely and retains 55%, the highest in the set — above the knit.

Measured input ranks with the cloth; the presets do not

Seven fabrics were draped on a self-built Cusick-style rig, three specimens × two faces each, and their drape coefficients compared against the same seven simulated from measured input and from library presets.

• Order holds, amount does not. Measured input reproduces the physical ranking at Spearman 0.964; the presets manage 0.286. On absolute error the two arms are nearly level — 0.106 against 0.116 — so measuring buys the ranking, not the value.
• Fabric by fabric: measured is closer on three, level on two and further on two, counting as level any fabric where the two arms differ by less than that fabric's own spread across its six readings (P06, 0.002 against a spread of 0.041; P15, 0.015 against 0.022). Both arms remain far from the cloth.
• The clearest case is a dyeing difference. P31 and P32 are one cloth in two colourways. Physically they differ by 0.202. The library preset gives them the same coefficient, 0.0001 apart — it cannot see them as different fabrics at all. Measured input separates them by 0.007, which is to say it does not recover the difference either.

Run the comparison on your own garment

DRAPE Helper is a Windows companion app that drives the same comparison on a garment file you supply, and writes the result as a PDF report.

Give it a CLO3D garment (.zprj) plus two fabrics (.zfab) — a baseline and a candidate — and it simulates both on your garment, measures the per-vertex difference, and reports it. Fast mode runs each arm once; full mode repeats three times so the run-to-run spread is visible next to the difference you are reading.

Installs and updates from the Microsoft Store. Windows 10 and 11. A local CLO3D installation is required — the app drives your copy of CLO3D, it does not replace it.

A delta is a match statement, not a score

Read a delta as: the library fabric is not a match to the fabric you are hoping to use. It is not a measure of how wrong CLO3D is, and a smaller number is not automatically the better garment.

• Every value is a within-CLO3D difference between a named library preset and the measured input — not an error against physical reality.
• Preset baselines are manually chosen CLO library references, not equivalent physical fabrics. There is no naturally correct baseline for the delta to be measured from.
• A measured value is a known reference point, not a target. Designers lower stiffness on purpose to exaggerate drape. Closer to measured is not automatically the better-looking garment — it is the position you can account for.
• A delta belongs to the garment it was measured on. What transfers between projects is the method, not the millimetres.
• A different CLO3D build is a different engine. Results from another version have to be reported separately, never merged into this record.

Where it is useful: before cutting a toile. If the digital fabric standing in for your cloth is 95 mm away from it on the garment you are about to sample, the fitting you approved on screen is not the fitting that arrives on the stand — and the cost of finding out is a cut length of cloth and a pattern revision.

Stated in full, on every page

This is an exploratory single-engine pilot by one MA student. The limits below are not disclaimers added at the end — they define what the numbers are allowed to mean.

• Single solver: CLO3D 2026.0.374, particle distance 10. No cross-engine comparison.
• Seven specimens, six specifications — P31 and P32 are the same specification dyed apart. n = 5 per reported test, warp and weft where the property is directional.
• ISO 139 atmospheric conditioning was not carried out. Declared as a departure, not omitted.
• Bending uses a non-standard cantilever procedure; shear is left at the engine default.
• Property-isolation evidence exists for P15 and P28 only. The other five fabrics have no repeat-run data, and say so where their numbers appear.
• Internal consistency checks (Spearman 0.98 / 1.00) are internal consistency only. They are not physical-truth validation.
• The physical rig is self-built and uncertified. v1 demonstrates one fabric intake channel end to end; the others are designed concepts.
• Provenance is exported exactly as it was declared. This deployment does not verify it.

The measured .zfab files are downloadable from each fabric card in the workspace, alongside the named library preset, both garment scenarios, the engine build and the fitting mode. The garment engineering files are not distributed, so the exact millimetre figures here cannot be reproduced — the method can.