Industries / Consumer goods / Jewelry
Design
Design is turning an idea for a piece into drawings, models and files that a workshop can make and a customer can see before it is made.
5
Milestones achieved, from 1581 to 2017.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1581
Jewel designs in print
After the Flemish designer Hans Collaert died in 1580, his sons and his publisher, Philips Galle, issued engravings of his jewelry designs. The first set, published in 1581 as "Monilium Bullarum Inauriumque Artificiocissimae Icones", has ten prints of pendants in strapwork set with pearls and mythological figures, by the Victoria and Albert Museum's record. A jewel's design now existed on paper, apart from any piece made from it.
Credit
Hans Collaert and Philips Galle; museums differ on which member of the Collaert family engraved the plates
1971
Copyright stops at the idea
Herbert Rosenthal Jewelry Corp. held a copyright on a gold pin shaped like a bee and set with jewels, and accused Edward and Lucy Kalpakian of copying it. The US Court of Appeals for the Ninth Circuit found that the Kalpakians had designed their own bees from nature and published works, and that the two pins were alike only as far as any jeweled bee must be. It held that when an "idea" and its "expression" cannot be separated, copying the expression is not barred.
LessonCopyright protects one particular rendering of a motif, not the motif, so a designer cannot own the idea of a jeweled bee.
Credit
Judge James R. Browning, writing for the court
2005
Printed patterns replace carved models
A modelmaker had always turned a designer's drawing into a master model by hand, as Steven Adler of Automated 3D Modeling and Teresa Fryé of TechForm Advanced Casting Technology put it in 2005. They wrote that over the previous five years CAD (computer-aided design) had become affordable, and designers were taking over the modelmaker's job with printers that build models layer by layer. One printer line traces to Sanders Prototype, formed in 1993 and renamed Solidscape in 2000. Adler and Fryé cast ring patterns from Solidscape, 3D Systems and EnvisionTEC printers in platinum and measured how each shrank or grew.
LessonPrinted pattern materials burn out and expand differently from wax, so casters had to work out new burnout settings for each material.
Credit
Steven Adler and Teresa Fryé (tests); Solidscape, 3D Systems and EnvisionTEC (printers)
2009
CAD reaches the store counter
Stuller, a Louisiana supplier of mountings, findings and stones to retail jewelers, partnered in January 2009 with Gemvision of Davenport, Iowa, which sold jewelry CAD/CAM software, milling machines and printers. The two said the move would make custom design with CAD more accessible to retail jewelers through Stuller's custom-order program, which began in 1995. Stuller later said it had acquired Gemvision in 2009, and in October 2015 it bought TDM Solutions of Barcelona, maker of the jewelry CAD program RhinoGold.
Credit
Gemvision (founder Jeff High) and Stuller (Matthew Stuller), as reported by JCK and National Jeweler
2017
Prongs drawn into the model
Many shops left the prongs and stone seats out of their CAD models for fear that thin cast prongs would be flawed, and let the stone setter build them by hand. Jerry Taylor of the metals supplier Hoover & Strong printed a test ring on an Asiga Pico2 printer with prongs 0.2 to 0.75 mm thick and 0.75 to 2 mm high, cast it and cut it open; all prongs were sound, with slight tapering on the thinnest that came from printing. By his account, top setters set fewer than 150 stones in an eight-hour day when they must form the prongs, and over 500 when prongs and seats come cast. Peter DiCristofaro of the Providence Jewelry Museum likewise credits printed waxes with under-bezels and seats, matched to calibrated stones, for rings that are easy to set.
LessonThe fear that kept prongs out of models did not hold up when tested.
Credit
Jerry Taylor, Hoover & Strong
Watch for
AI agents draft castable rings
Pencil says its AI design agent turns plain-language requests into rendered, parametric CAD designs meant for printing and casting, not just images that still need a CAD designer. Users can set limits such as prong diameters, minimum band widths and polishing allowances, and ask for exact changes, such as moving a stone 1.25 mm. Ring design is available first.
Achieved whensomeone outside the company prints and casts a sample of agent-made rings and publishes how many needed CAD rework before casting and setting.
Claimed by
Pencil, May 2026
Try-on that keeps true size
Virtual try-on puts a picture of a piece onto a photo of the shopper. JewelTry is a diffusion model, a kind of image generator, that takes a product's real dimensions as an input so a ring or earring is drawn at the right scale on the wearer. Its authors also introduced JVTO-Bench, a test set with real product-scale annotations across four jewelry categories, and report that JewelTry balances image quality with scale accuracy on it.
Achieved whensomeone outside the team measures its size error against photographs of the same pieces actually worn, or a retailer uses it and publishes measured error.
Claimed by
Xinlei Niu, Peixia Li, Jun Wang, Chenchen Xu, Jiayu Yang and Jing Zhang, September 2026
Open
01
Castability test for AI rings
A public benchmark of about 100 written ring briefs (finger size, metal, stone shapes and sizes, setting style), the files several AI design tools return for them, and the metal results. Score every file with open rule checkers for closed solids, inner diameter, band and prong thickness and stone-seat fit, then print and cast a sample to learn whether a checker's pass or fail predicts a sound casting.
The briefCredit · unclaimed02
True-size try-on audit
A public set of photographs of real pieces worn on real hands, ears and necks, each piece measured with calipers and photographed beside a printed scale card, paired with images of the same pieces produced by virtual try-on tools. Measure each tool's size error in millimeters and percent.
The briefCredit · unclaimed
Open 01 · the brief
Castability test for AI rings
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public benchmark of about 100 written ring briefs (finger size, metal, stone shapes and sizes, setting style), the files several AI design tools return for them, and the metal results. Score every file with open rule checkers for closed solids, inner diameter, band and prong thickness and stone-seat fit, then print and cast a sample to learn whether a checker's pass or fail predicts a sound casting.
Why it is still open
Pencil says its agent's designs are ready for printing and casting but has published no outside test. Meshy, a text-to-3D service, tells jewelers to bring its models into CAD to finish when they need production precision. Open checkers now exist without casting tests behind them: Paloan Jewelry CAD, posted on GitHub under the MIT license in October 2026, has a rules engine that passes, warns or fails shank thickness, prong diameter, seat depth and other limits, and JewelryAI Studio (October 2026) builds solitaire rings in CadQuery with what it calls preliminary checks, "not an automatic manufacturing approval system". The jewelry datasets on Hugging Face are photographs for training image generators, such as one with about 6,100 images; the nearest published work estimates metal loss in casting from CAD file attributes, on a small set of rings (Jain, Jain and Mane, 2023). Searches of Hugging Face, arXiv, Crossref and GitHub in November 2026 found no benchmark that scores AI-made jewelry CAD against cast results.
jewellerymonthly.co.ukmeshy.aigithub.comgithub.comhuggingface.coarxiv.org
Built with what exists
A model such as Claude Fable 5.2 writing CAD scripts in CadQuery or Rhino's Python (see the Note "Three working devices"), plus text-to-3D services and jewelry CAD agents as other entrants; Paloan's open manufacturing checker and geometry checks in Python (trimesh, Open CASCADE); a desktop resin printer with castable resin, such as the Asiga Pico2 used in Taylor's 2017 prong test; and a commercial casting service for the metal step.
First step
Write 20 ring briefs with a bench jeweler, have one AI system answer them, and run Paloan's checker on the files it returns.
Done looks like
100 briefs and scores for at least three AI systems, published with the scoring code; and 30 rings printed and cast in which the checker's pass or fail matches the casting outcome (sound or flawed, as judged by a bench jeweler) in at least 27.
Open 02 · the brief
True-size try-on audit
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of photographs of real pieces worn on real hands, ears and necks, each piece measured with calipers and photographed beside a printed scale card, paired with images of the same pieces produced by virtual try-on tools. Measure each tool's size error in millimeters and percent.
Why it is still open
JewelTry's authors measure scale on their own benchmark of generated images. A 2025 ring-size estimation system built on MediaPipe hand landmarks gives no measured error in its abstract, and GlamTry (2024) reports better placement, not size. Searches of trade and academic sources in November 2026 found no independent test of commercial jewelry try-on tools against photographs of the same piece worn.
Built with what exists
A phone camera on a stand, digital calipers, a printed scale card, volunteers who consent in writing to their hand and ear photos being published, open hand-landmark models (MediaPipe Hand Landmarker), and the try-on tools on retailers' websites.
First step
Photograph 10 rings on 5 volunteers' hands beside a scale card, then put the same rings through two retailers' try-on tools and compare widths in millimeters.
Done looks like
50 pieces photographed on 10 wearers, with each tool's median size error and its spread reported, and the photo set released for others to test new tools.