AI Jewellery Component Detection & Auto-Split
Upload one image. Facetra detects every chaton, prong, jaali panel, meena inlay, drop and chain — and splits them into individually addressable regions ready for casting, refinement or 3D generation.
Every serious jewellery piece is really an assembly: a Rani Haar is seven panels, thirty chatons, twelve meena inlays, four ghungroo drops and a chain — thirty-plus distinct components that a karigar or CAD operator has to model one-by-one. Traditionally, decomposing that into a workable component list is a 3-6 hour manual job in Photoshop or Rhino, or an eyeballed guess from the render.
Facetra Auto-Split does it in 20 seconds. Upload a single image (sketch, render, or DSLR photo) and the engine detects each component, assigns it a semantic label (chaton / prong / jaali / meena / drop / chain), and outputs individual PNG masks + a JSON manifest. From there you can regenerate just one component, feed each component into 3D CAD generation, or hand the CAD bench a component-by-component build list.
What is AI component detection for jewellery?
AI jewellery component detection is a segmentation workflow that identifies each meaningful sub-element of a jewellery piece and separates it into an individually addressable region. Unlike generic image segmentation (which detects "jewellery" as a single blob), Facetra's engine is jewellery-native — it distinguishes a chaton from a bezel, a prong from a granulation bead, meenakari from enamel-paint, and a Kundan-set edge from a modern pave halo. The output feeds three downstream workflows: (1) targeted refinement (regenerate one component without touching the rest), (2) component-level 3D generation (each part becomes its own STL for the CAD bench), and (3) manufacturability planning (the karigar gets a component-by-component build list).
What Facetra Auto-Split delivers
- Jewellery-native semantic labels — chaton / prong / jaali / meena / granulation / drop / chain / clasp / rivet / halo / gallery / shank
- Style-aware — the same silhouette gets labelled correctly whether it's Kundan, Jadau, Polki, Diamond, Temple or Meenakari
- Individual PNG masks per component + JSON manifest with coordinates + bounding boxes
- Targeted refinement — pick one component, describe a change ("swap the emerald centre for a ruby"), only that region re-renders
- Component-level 3D pipeline — each detected part can be fed into 3D STL / GLB generation independently
- Manufacturability output — component count, dominant material per part, and a build-order suggestion for the karigar
The Facetra workflow (six steps, one afternoon)
- Upload one image — sketch, Facetra render or DSLR photo.
- Optional: tag the style (Kundan / Jadau / Polki / Diamond / Temple / Meenakari) so the labeller uses the right vocabulary.
- Facetra runs component detection in ~20 seconds.
- Review the labelled components in the split viewer — reorder, rename, or merge if needed.
- Choose downstream action: targeted refinement, per-component 3D generation, or export component build list.
- Download the ZIP of PNG masks + JSON manifest, or push directly into the 3D pipeline.
Sample prompts to copy
Paste any of these prompts into Facetra Studio to see the style in action. Adjust weights (heavy / light), stone colours and motifs freely.
How much does it cost?
Credit-based pricing. Studio render 1 credit per variant, multi-view sheet 4 credits, CAD-ready view 1 credit, 3D mesh 15 credits (Standard) or 25 credits (HD). Credit packs start at ₹1,500 (Starter, 100 credits, ₹15/cr) and go down to ₹12/credit on the Atelier pack (1,300 credits at ₹15,600). New signups get 10 free credits.
Frequently asked questions
How is this different from generic image segmentation?
Generic tools (Segment Anything, Rembg, etc.) treat jewellery as one blob or produce anatomical parts ("top half", "bottom half"). Facetra Auto-Split understands jewellery vocabulary — a chaton is not the same as a bezel, a jaali panel is not the same as a meenakari inlay. That semantic layer is what makes the output usable by a karigar or CAD operator.
Can Auto-Split feed into 3D CAD generation?
Yes — that's the primary downstream use. Each detected component becomes its own STL / GLB / OBJ mesh via Facetra's 3D pipeline, giving your CAD bench a component-by-component library instead of a monolithic mesh they have to slice manually.
Does it handle heritage silhouettes like Aad, Vanki, Oddiyanam?
Yes. The engine is trained on Indian bridal silhouettes and knows the anatomy of aad panels, vanki inverted-V armbands, oddiyanam waist belts, bajubandh armlets, hathphool hand ornaments, maang tikka and nath. Regional variants can be handled by uploading a reference photo.
How much does Auto-Split cost per image?
2 credits (~₹24) per image at the Atelier pack rate. Bulk-mode discounts apply from 100+ images per submission.
Can I re-render just one component after Auto-Split?
Yes. Pick any component from the split viewer, describe the change ("deeper green meenakari on this panel", "swap this chaton to oval Polki"), and only that region re-renders — the rest of the piece stays pixel-faithful. This is the fastest way to iterate with a client without re-rendering the whole ₹4L Rani Haar.
10 free credits. No card required.
Enough to render 15 variations, or one full CAD kit — see the output before committing to a pack.
Create free account