A furniture retailer wanted AR room-placement for every SKU, but commissioning 3D scans was $80–200 per item. We built a TRELLIS-based image-to-3D pipeline that generates AR-grade GLB meshes from a single product photo in 90 seconds.
The retailer's app has “place this in your room” AR - but only on the 12% of SKUs they'd paid to 3D-scan. The other 88% just showed a flat image. Conversion on AR-enabled items was 2.3× higher than non-AR, so this was a real revenue hole. Commissioning 3D scans at scale was a multi-million-dollar non-starter.
Benchmarked Microsoft TRELLIS, InstantMesh, and Wonder3D on 200 reference products spanning sofas, lamps, vases, and small accessories. TRELLIS-image-large won on mesh quality and PBR material fidelity.
Built a preprocessing step that crops, background-removes, and category-classifies the product photo - then routes to the right TRELLIS prompt and material preset based on category (matte / fabric / metal / glass).
When multiple catalogue angles exist for an SKU, we condition TRELLIS on the front shot and refine the mesh against a side and rear shot. Cuts geometric error by ~40% on asymmetric items.
Generated meshes auto-decimated to AR-app polycount budgets (15k tris for hero items, 6k for accessories), baked PBR materials, and stamped with SKU metadata.
Async job queue on a single DGX Spark - 12 GB VRAM per render - pushing GLBs into the retailer's existing CDN. Models that pass QC are cached forever. ~6k SKUs generated in the first month.
“Every SKU now has AR. We stopped paying per-scan and the conversion lift paid for the whole project in two months.”
- Head of Product, furniture retailer (name withheld)
A 30-minute call. We'll tell you whether we can help - and if not, who can.