Image-to-3D for product design and Japanese e-commerce
Japanese e-commerce has been slower to adopt 3D product previews than some other markets, partly because producing a 3D model from a physical product used to require a photogrammetry rig or a skilled modeller. Image-to-3D generation changes that calculation significantly.
What image-to-3D needs from your reference photo
The generator works best with a single clean photo taken against a neutral background with even lighting. A white or light grey background removes shadow ambiguity. The product should fill most of the frame and be photographed from a three-quarter angle so the generator can infer depth.
Highly reflective surfaces like polished chrome or glass are harder to reconstruct accurately from a single image because the reflections carry no depth information. For these materials, photograph the product with a polarising filter or use a matte spray to reduce reflections temporarily.
Refining the reconstruction through chat
The first reconstruction from a single image is rarely perfect. Common issues are a slightly incorrect base shape, missing detail on the back face, or an inaccurate texture on a label or pattern. All of these can be corrected through the chat interface without starting over.
For the back face, type 'the back of the product is flat with a single USB-C port centred at the bottom' and the agent will update the geometry. For label text, it is usually faster to apply the label as a decal in your target application rather than trying to generate it accurately through the AI.
Exporting USDZ for Safari AR Quick Look
Once you are happy with the model, export as USDZ from the export panel. Set the scale in centimetres to match the real product dimensions. Safari's AR Quick Look uses the scale metadata to place the product at the correct physical size in the user's room.
Test the USDZ on an iPhone before publishing. Open the file in Files app and tap it to launch AR Quick Look. Check that the scale is correct, the materials look right under room lighting, and the model loads in under three seconds on a mid-range device. If it loads slowly, reduce the polygon count through the chat interface before re-exporting.
A USDZ product preview on a Japanese e-commerce page is still uncommon enough to be a genuine differentiator. The technical barrier is lower than it has ever been.