The three forms: viewers, AR, and renders
"3D furniture visualization" covers three distinct deliverables, and most confusion about cost and process comes from not separating them. An interactive 3D viewer lets a shopper rotate and zoom a product on the page itself — it replaces the two or three angles a photographer chose with the full form, from any angle, in the browser. AR placement ("view in your room") goes further: it takes that same 3D model and, using the camera on the shopper's phone, places it at true scale on their actual floor, next to their actual sofa or doorway, with no app download on modern iOS and Android. A photorealistic render is different again — a single high-quality still image generated from 3D source data, used for marketing and catalog photography rather than interactive shopping. Viewers and AR share one underlying model and are usually bought together; renders are a separate output that can reuse the same source geometry but serve a different job.
How 3D models get made
There are three practical paths to a furniture 3D model. From CAD data: if a manufacturer already has engineering-grade CAD files, those can be converted to a web-ready model — fastest and most exact, but most furniture retailers never receive CAD from their suppliers. From photos: image-to-3D generation (photogrammetry-style or AI-assisted) builds a model from a handful of clean product photos plus merchant-entered dimensions, then a human reviewer checks silhouette, scale, and material before anything ships — this is the practical default for stores without CAD, because photos already exist for the product listing. Hand-modeled: a 3D artist builds the model manually in modeling software, matched against reference photos and measurements — highest control over fidelity, but the slowest and most expensive path, typically reserved for a small number of hero SKUs. Photo-based generation with human review is the middle ground most stores land on: near-CAD-level shopping accuracy without needing CAD to exist in the first place.
What it changes commercially
The commercial case rests on three levers, each with furniture-specific evidence. Conversion: furniture retailer DFS, the UK's largest furniture retailer, saw shoppers who interacted with 3D and AR convert 112% more often and generate 106% more revenue per visit than those who didn't, for a reported 22x return on the AR investment (Vertebrae/DFS, 2020); Houzz found shoppers who used its AR "View in My Room" feature were 11 times more likely to complete a purchase (Houzz, Internet Retailer Conference, 2018). Returns: size or space mismatch alone drives 58% of furniture returns (RocketReturns, 2025), and Shopify reports that 3D/AR product content is associated with return rates falling by up to 40% when the model sets accurate size expectations before checkout; Macy's reported 25% fewer returns alongside a 60% larger average basket in a VR/AR furniture pilot. Engagement: 3D and AR give shoppers a reason to spend more time with a product page before deciding, which is itself a leading signal — pages with strong viewer and AR engagement but flat conversion usually point at a missing dimension table or fabric close-up, not a 3D problem.
File formats and delivery
In plain terms: GLB is the standard 3D file format for the web and Android AR (Google's Scene Viewer opens GLB directly from the browser). USDZ is Apple's format for iOS AR — iPhone's Quick Look viewer requires USDZ and will not open a GLB, so a model destined for iPhone AR needs both files exported from the same source. Compression matters because an uncompressed GLB out of most 3D tools can run 20-50 MB, which stalls on mobile connections; Draco geometry compression plus compressed textures typically bring a furniture model down to a few megabytes with no visible quality loss. A poster image — a static picture shown while the 3D file streams in — keeps the page from looking broken on a slow connection. Delivery over a CDN, with correct CORS headers so the viewer can load the files cross-origin, is what keeps page load fast at catalog scale rather than per-product.
Build vs buy
Three practical routes exist. In-house 3D team: full control and no per-model vendor cost, but it means hiring or training 3D artists, building and maintaining a hosting and viewer pipeline, and owning USDZ conversion and compression — realistic mainly for retailers with sustained volume across hundreds of SKUs. Modeling studios: you send photos or specs, a studio delivers model files, and you're then responsible for hosting, the viewer, AR packaging, and updates when products change — publishing quoted a wide per-model price range for exactly this reason (see the cost guide for the current breakdown). Done-for-you services: photos and dimensions in, a live hosted 3D/AR product page out, with hosting, viewer, AR packaging, human review, and analytics included rather than left to you — the tradeoff is less raw control over the pipeline in exchange for not building one. None of these is universally right; the deciding factor is usually catalog size and whether a team already exists to own an ongoing pipeline versus a single product experience.
Getting started
Start with a pilot, not the full catalog: 10-25 hero SKUs where shoppers most often hesitate on size or fit — sofas, sectionals, dining tables, and large case goods are the usual candidates, since a sizing mistake there is the most expensive to ship back. Follow a basic photo checklist for each product (straight front view, side or three-quarter view, back view, and a material detail shot, with real width/height/depth recorded) so the source photos support an accurate model rather than an approximate one. Once the pilot is live, measure it rather than assume it worked: AR launches, 3D viewer interactions, and clicks back to the store, tracked per product, tell you which SKUs earned real engagement and are worth expanding to next, and which ones need a better dimension table or fabric shot before more 3D spend is justified.
