Identifying 700+ Luxury Fabrics — Without a Single Tag or Mark

For a luxury fashion house, the fabric is the product. Hundreds of bolts — silks, jacquards, brocades, fine wools — many of them visually near-identical, each with its own composition, supplier and price. The team could tell them apart by eye and by touch, but that knowledge lived in a handful of senior people, and finding the right bolt for a client could take minutes of searching through memory and paperwork.

The challenge

With more than 700 fabric varieties in active inventory, identification had quietly become the bottleneck in everything from client consultations to reordering.

• check_circleIdentification depended on a few expert staff — onboarding a new team member took months.
• check_circleVisually similar fabrics were easily confused, leading to wrong pulls and unnecessary reorders.
• check_circleFinding a specific fabric meant manual lookup across registers, memory and paperwork.

Why marking the fabric was off the table

The obvious answer — tag every bolt — was a non-starter for a clientele this discerning.

• check_circleRFID tags, woven labels and adhesive stickers attach to or pierce the cloth — unacceptable on luxury textiles.
• check_circlePen marks, stamps or any residue permanently devalue the fabric.
• check_circleBolt-end barcodes get cut off or lost as the fabric is sold down.
• check_circleThe clientele will simply not accept any visible alteration to a premium material.

On fabric this fine, the identifier cannot touch the cloth. The only acceptable tag is no tag at all.

What we built

PetalKube built a non-contact, vision-only identification system. A camera — a fixed capture station or a staff phone — photographs the fabric. An object-detection model isolates the cloth, and a fine-grained classifier matches it against all 700 varieties, returning the exact SKU with its composition, supplier and stock location in under two seconds. Nothing is ever attached to the fabric.

How it works

• looks_oneCapture — staff frame the fabric at a station or with a phone; no special lighting rig required.
• looks_twoDetect — the model locates the fabric region and ignores background, hands and packaging.
• looks_3Classify — a fine-grained model trained on all 700 varieties reads weave, pattern, colour and texture.
• looks_4Match — the system returns the top match with a confidence score, plus close alternatives.
• looks_5Resolve — the SKU pulls composition, supplier, price and location from inventory and logs the lookup.
• add_circleLearn — new fabrics are added to the model continuously as the collection grows.

The outcome

The expertise that once lived in a few people now lives in the system, available to anyone on the floor. Mispicks and unnecessary reorders dropped, client-facing staff move faster, and — most important to this client — every bolt remains exactly as pristine as the day it arrived.