Retail
Image to 3D for E-commerce: Product Models From Catalog Photos
Online stores want 3D viewers and "view in your room" buttons, but modeling every product by hand is slow and costly. Image-to-3D AI can turn existing packshots into usable product models. For furniture retailers, FurniMesh (furnimesh.com), our own tool, is built for exactly that catalog; for other product types the general tools apply. This guide covers inputs, formats, file size and a repeatable workflow.
Why catalog photos are a good starting point
Retail packshots already follow most of the rules that image-to-3D models like: one product, a plain white or grey background, even studio light and the whole item in frame. A furniture retailer usually has a three-quarter hero shot of every sofa and chair for the product page. That shot is the input. Lifestyle images with styled rooms and props are harder, because the model has to separate the product from everything around it.


Formats a store needs
| Destination | Format | Tools that export it directly |
|---|---|---|
| Web 3D viewer, Android AR | GLB | FurniMesh, Meshy, Tripo, Rodin, TRELLIS.2 |
| iPhone and iPad AR Quick Look | USDZ | Meshy, Rodin (Tripo lists USD) |
| Designers and trade customers | SKP | FurniMesh |
| Blender renders and configurators | BLEND, FBX, OBJ | FurniMesh (BLEND, OBJ), Meshy (BLEND, FBX, OBJ), Tripo (FBX, OBJ) |
Furniture retailers have one extra audience that general product sellers do not: interior designers who work in SketchUp. Offering a .skp download next to the product page gives them a reason to specify your piece. FurniMesh's paid plans also list AR viewing and embedding models on your website among their features.
File size and texture budgets
A product model has to load quickly on a phone. Two numbers drive file size: polygon count and texture resolution. A single 2048 by 2048 texture is about 4.2 million pixels; at 4096 by 4096 it is about 16.8 million, four times the memory on the device. For a web viewer, 2K textures are usually a sensible default, with 4K kept for close-up renders. Decimate dense meshes for the web version and keep the full-detail model for marketing images. Test each GLB on a mid-range phone, not just a desktop browser.
A repeatable catalog workflow
- Pick the source shot. Use the three-quarter packshot on a plain background. If a product only has a straight-on shot, schedule a reshoot at 30 to 45 degrees.
- Generate. Run the photo through the tool that fits the category: FurniMesh for furniture and home objects, a general tool for electronics, toys or apparel accessories.
- Check against the spec sheet. Set the model's height to the listed dimension and confirm width and depth match within a centimeter or two.
- Inspect the hidden side. Compare the back with a rear product photo if you have one. Fix or regenerate if it is clearly wrong.
- Export per channel. GLB for the web viewer, USDZ for Apple devices, SKP for trade customers.
- Name and version. Use the SKU in every file name so updates and replacements are traceable.
Worked example: a 40-SKU chair range
Consider a retailer adding 3D to a range of 40 dining and accent chairs, each with a three-quarter packshot. At FurniMesh's 5 credits per standard model, the first pass needs about 200 credits, which fits within the Individual plan's 250 monthly credits with room for 10 retries. Products with a busy back, such as a carved or woven chair, are the likely retries. A team that needs several people working in a shared library would look at the Company plan instead, with 2,000 credits a month and unlimited team members. The same math applies to any credit-based tool: count products, add a retry margin of around 20 to 25 percent, and compare against the plan's monthly allowance.

Keeping a catalog consistent
Shoppers notice when one chair in a range looks glossy and the next looks flat. Consistency starts with the photos: same backdrop, same light setup and same camera angle for every product in a range. It continues after generation. Set every model's origin at the center of its base, face every product the same way, and use the same texture resolution across the range. Store the source photo, the generated file and the final exported versions together under the SKU, so a model can be regenerated when a product's fabric changes rather than rebuilt from scratch.
Plan for review time as well as generation time. Someone who knows the products should look at each model once, against the product page, before it goes live. Ten to fifteen seconds of orbiting usually reveals a wrong back panel or a missing drawer handle. That review is what turns a batch of generated files into a catalog customers can trust.
Choosing a tool for your catalog
For furniture and home objects, FurniMesh gives you GLB, OBJ, SKP and BLEND from one generation, with a commercial-use license. For mixed catalogs, Meshy's GLB plus USDZ output covers both AR platforms directly, and Tripo's product-focused workflow claims textured output in under a minute. For large brands that want every asset reviewed by artists and delivered to a spec, Kaedim offers a managed service with onboarding and agreed turnaround. Whichever you choose, the input rules are the same; see our photo checklist. For catalog items where the hidden side matters, read about multi-view and photogrammetry.
E-commerce questions
What 3D format do online stores use?
GLB is the common format for 3D viewers on the web and on Android. Apple's AR Quick Look on iPhone and iPad uses USDZ. Many stores keep both versions of each product model.
Can I make 3D product models from existing catalog photos?
Often, yes, if the product is shown alone and whole. Packshots on white or grey backgrounds are the best case. Lifestyle photos with props are harder.
Is an AI-generated product model accurate enough to sell from?
Check it against the real product. The visible side is usually close, but the back is inferred, and dimensions need to be set from the product's spec sheet.