Use case · AI content creation

AI and 3D product visualisation, from CAD file to AR.

3D and AI solve different halves of product visualisation. A 3D model built from CAD gives exact geometry for every angle, colourway and AR view; generative AI supplies scenes, lighting and context around it quickly. The work is in the pipeline between them: turning heavy engineering files into light real-time models, publishing the right AR formats (glTF/GLB and USDZ), and keeping renders consistent with photography.

From engineering CAD to a usable model

Engineering CAD is built for manufacturing, not for images. Files arrive as STEP or native formats describing precise surfaces, often with internal parts nobody will see and no materials. Turning them into a visual asset means tessellating surfaces into polygons, removing hidden parts, reducing polygon counts for real-time use, unwrapping UVs and assigning physically based materials with base colour, metallic, roughness and normal maps. Keep two outputs: a high-detail model for offline renders and a light model for the web and AR. Image-to-3D generation can produce plausible shapes from photos, but not engineering-accurate ones, so use it for props and scenery, not for the product.

Configurators and material variants

A configurator lets buyers change colour, material or components and see the result, which only works if every option is modelled and materialised accurately. glTF 2.0, the Khronos Group standard for real-time 3D, has a materials-variants extension (KHR_materials_variants) so one file can carry several finishes without duplicating geometry. Match each digital material to a physical swatch under controlled light, because a leather or anodised finish that is slightly off becomes a returns problem. Generated imagery then extends the configurator into marketing: the same configured product placed in rooms, streets or seasonal scenes, rendered from the model so geometry and finish stay exact.

AR formats: glTF/GLB and USDZ

Augmented reality on phones splits by platform. Apple's AR Quick Look uses USDZ; Android's Scene Viewer and most web viewers use glTF, usually packaged as a single binary GLB file. Most brands publish both from the same source model. Keep files light enough to load on a mobile connection: compress meshes with Draco, compress textures with KTX 2.0, and test real-world scale, because a sofa that appears at the wrong size in someone's living room defeats the point of AR. Check lighting and material appearance on actual devices, not only in the authoring tool, since each viewer renders materials slightly differently.

Keeping renders and photography consistent

Most catalogues mix photography, 3D renders and generated scenes, and customers notice when the same product looks different between them. Match camera focal length and height, lighting direction and colour temperature, and keep one colour-managed workflow from render to final file. Approve all three against the physical product. The studio's product-lock keeps a real product pixel-accurate in geometry, labels and colours while the scene around it changes, and saved elements reuse the same product, set and style across stills, video and formats, so generated scenes start from the product as photographed or rendered rather than a reinterpretation of it.

Updated 25 September 2026

Questions

AI and 3D product visualisation, answered.

Can AI create a 3D model from product photos?

Image-to-3D models can produce a plausible shape from a few photos, which is useful for props, previews and scenery. They do not reproduce exact dimensions, fine details or accurate materials, so products that customers configure, measure or place in their homes should be modelled from CAD or scanned properly.

Which 3D formats do we need for AR on product pages?

Usually two, from one source model: USDZ for Apple devices, which open it in AR Quick Look, and glTF/GLB for Android and web viewers. Compress both for mobile loading, and check on real devices that the product appears at its true size.

Should we use 3D renders or generated images for our catalogue?

Both, for different jobs. 3D renders give exact geometry and every variant from one model; generative AI adds scenes, context and seasonal settings quickly. The combination works when the product itself stays locked, whether it comes from a render or a photograph, and only the surroundings are generated.

Get started

See your AI studio generating this week.

Thirty minutes. A live studio in your colours, your brand hub loaded, a real campaign brief.

See plans