Creating a 3D digital twin of your products

A 3D digital twin of a product is an accurate virtual model of it, with the right geometry, materials, artwork and variants, kept current as the product changes. Images, videos, interactive views and AR experiences can then be generated from the twin instead of being photographed, for every angle, market and pack version.

The term comes from engineering, where a digital twin is a virtual counterpart linked to a physical thing. That origin explains what makes a product twin useful for marketing. It is only as good as its fidelity to the physical product and its currency with the product’s latest versions. So creating one is mostly a question of inputs, materials, artwork, variants and validation, and much less a question of which modelling software to use. This guide covers what a 3D digital twin is, examples across industries, how to create one step by step, what “AI digital twin” means, and what product twins cost. For Grip’s own twin production, see the Digital Twin Factory.

What is a 3D digital twin?

The idea goes back to a presentation Michael Grieves gave at the University of Michigan in December 2002, on setting up a centre for product lifecycle management. His slide, titled “Conceptual Ideal for PLM”, showed a physical product and a virtual one linked by data flowing between them. According to Grieves and John Vickers, the name “digital twin” came from Vickers, of NASA, and was first attached to the model in Grieves’s 2011 book Virtually Perfect.

NASA’s 2010 technology roadmaps described a digital twin as an integrated simulation of an as-built vehicle that uses physical models, sensor updates and fleet history to mirror the life of its flying twin. The US National Academies’ 2024 definition keeps that link at the centre: a digital twin mimics a system’s structure, context and behaviour, is updated dynamically from its physical counterpart, and informs decisions.

A product twin for marketing has no sensor feed, so it is closer to what Grieves and Vickers call a digital twin prototype: the information needed to describe a product and make it. Its link to the physical product runs through versions rather than live data.

Digital twin examples

Digital twins are used across industries, at very different scales:

• Factories. BMW has digital twins of more than 30 production sites in its Virtual Factory, whose simulation application is based on NVIDIA Omniverse. It says a collision check that took almost four weeks of physical testing now takes three days virtually, and projects up to 30% lower production planning costs.
• Cities. Virtual Singapore, led by the country’s National Research Foundation with the Singapore Land Authority and GovTech, is a 3D model of the city enriched with live data and designed to study solar potential, flooding, wind and pollution.
• Healthcare. Dassault Systèmes’ Living Heart Project has built a virtual model of the human heart, used in research on testing medical devices in simulation.
• Cars. DENZA, the joint venture of BYD and Mercedes-Benz, worked with WPP to build its N7 car configurator from a single physically accurate twin, built in OpenUSD from CAD data, that holds every variant, so no images have to be made in advance. NVIDIA notes that traditional configurators need hundreds of thousands of them.
• Fashion. Zalando and the 3D capture company ALLSIDES scanned 10,000 footwear SKUs in 2025, at about seven minutes a scan, with 35,000 more planned for 2026, according to an NVIDIA case study.
• Beauty. L’Oréal said in June 2025 that its CREAITECH content platform already uses 3D product imagery for campaigns. For more examples in beauty, see 5 digital twin use cases in beauty.

How to create a 3D digital twin of a product

1. Gather the inputs

A product twin draws on several sources, and none of them is enough alone:

• CAD and engineering data for the shape and dimensions of the pack or product.
• Packaging dielines and approved artwork, usually as print-ready PDFs, for labels and decoration.
• Physical samples of the current product, for reference and validation.
• Reference photography under known lighting.
• Material samples or measurements for colour, gloss, transparency and texture.
• A variant list, covering sizes, colours, flavours, languages and pack versions, tied to each product’s GTIN.

CAD describes the product as designed, not as made. In one comparison published by the visualization software company Eclat Digital, a bottle modelled from CAD missed the flat top of the real bottle, and variations in production wall thickness, combined with translucency, changed its colour. Physical samples are what catch those differences.

2. Build the geometry

The shape comes from CAD where it exists, or from scanning where it doesn’t. Photogrammetry, structured light and laser scanning all depend on surfaces that reflect light diffusely, so glass and clear plastics are hard to capture, and the usual workaround is to coat the object in powder before scanning. Newer neural methods such as NeRF (2020) and 3D Gaussian splatting (2023), the second of which can display captured scenes in real time, record appearance well but store the lighting of the capture, so relighting a product for a new scene needs extra work. For marketing use, the geometry also needs true-to-life scale and clean surfaces where labels will sit.

3. Define the materials

Materials decide whether a twin looks like the product or like a model of it. Most production twins use physically based (PBR) materials, an approach shaped by work at Disney’s animation studio, whose 2012 shading model was informed by a study of measured material data. Physically based materials are meant to behave consistently across lighting conditions. The glTF standard’s PBR parameters give a sense of what is described: base colour, metalness, roughness, surface detail and emission, with extensions for clearcoat, transmission, sheen and volume. Where colour and finish matter most, such as car paint, cosmetics and premium packaging, materials can be measured rather than estimated, for example with X-Rite’s AxF format for measured appearance data.

4. Apply the artwork and decoration

Labels, print, foil, embossing and other decoration come from the approved artwork files, converted into textures with colour management so that the brand colours match the printed pack. This step ties the twin to the artwork system: when the artwork changes, the twin changes with it, which is what keeps product images current. For why that matters on retailer sites, see packshots at scale.

5. Model the variants and versions

A twin becomes efficient when one model holds the whole range: every size, colour, flavour, language and pack version, each tied to its GTIN and pack version. DENZA’s configurator shows the principle at scale, with one twin holding every variant where a traditional configurator needs hundreds of thousands of pre-made images. Without variants and versions, a brand ends up with a library of separate models that drift out of date one by one.

6. Validate against the physical product

A twin should be checked against the physical product before it is used, by comparing generated images with photographs taken under the same controlled light and from matched camera positions. Colour can be measured as a colour difference (ΔE). In Eclat Digital’s work on car paint, a difference of 1.0 wasn’t noticeable and 2.0 was barely visible, while glossy finishes were harder to match in direct sunlight. Published standards for validating product twins against physical products are scarce, so set an internal one: the views to compare, the lighting, the colour tolerance and who signs off.

7. Choose the format and keep the twin current

OpenUSD is widely used for production twins. The Alliance for OpenUSD, founded in 2023 by Pixar, Adobe, Apple, Autodesk and NVIDIA, published the first OpenUSD core specification in December 2025. It covers how scenes are composed and the USD file formats, and leaves geometry, materials and physics to later standards, so material interchange is not yet fully standardized. glTF 2.0, published as ISO/IEC 12113 in 2022, is widely used for web viewers and AR. A practical setup is a master twin in OpenUSD, with glTF or USDZ exports for specific channels. Whatever the format, assign an owner for each twin and update it whenever the artwork, formula or pack changes.

AI digital twins: what the term means

“AI digital twin” is used in at least four ways:

• AI replicas of people. H&M said in 2025 that it was creating AI twins of 30 models; the models reportedly keep the rights to their twins, and the images carry an AI watermark.
• Industrial twins for training AI. NVIDIA’s Mega blueprint, announced in January 2025, uses factory twins to train and test fleets of robots.
• AI that generates 3D models. Research models such as Microsoft’s TRELLIS.2, which Microsoft describes as purely a research project, and Meta’s SAM 3D, released in November 2025, generate 3D objects from images. Meta notes that SAM 3D’s moderate resolution limits fine detail, which is where brand accuracy lives: labels, logos and exact proportions.
• Twins that ground generative AI. Here the product twin keeps the product exact while generative AI creates the scene around it. NVIDIA published an Omniverse blueprint for this in November 2024, which uses masking to keep the product untouched while a generative model creates the scene around it, and named Grip among the companies adopting its Omniverse Blueprints.

For marketing, the fourth use is the one that matters most. It combines the accuracy of the twin with the flexibility of generative AI. AI product photography compares it with the other ways of creating product images.

Digital twin marketing: what product twins are used for

Once a twin exists, it can supply content for almost every channel: e-commerce packshots and alternate views, dynamic creative, social content, AR and VR, virtual showrooms, interactive configurators, TV commercials and out-of-home advertising. The same twin supplies all of them, so the product looks the same everywhere, and a pack change is made once, in the twin.

How much does a 3D product twin cost?

Independent cost data is scarce. Shopify put a simple 3D product model at $40 to $200 in 2020, with complex models costing more than $1,000, depending on the product’s form, materials and the reference material available. The main cost factors are:

• Shape and construction. Simple packs cost less than products with moving parts or complex geometry.
• Materials. Glass, liquids, metal, foil and fabrics take more work to get right.
• Decoration. Detailed labels, embossing and multiple languages add work per variant.
• Variants. The first variant costs the most, and each additional one costs less when the twin is built to hold them.
• Quality level. A twin for a web viewer needs less detail than one used for a hero campaign or TV.

Compare that cost with the shoots a twin replaces over its life, including reshoots for every pack change, new angle and market.

What it looks like in practice

Grip’s Digital Twin Factory produces tens of thousands of product twins a year using a Six Sigma process, and reports a 96% first-time-right rate within twelve months. Its twins are built in OpenUSD with high-polygon, true-to-scale geometry, physically based materials and colour-managed conversion of packaging PDFs into textures, and the factory works with NVIDIA Omniverse. Brands it lists among its customers include L’Oréal, LVMH and Shiseido in beauty and luxury, Moët Hennessy, Diageo and Pernod Ricard in beverages, and Kellanova, Heinz and Campbell in food. Grip’s own test for a twin is whether it can be used for consumer-facing content such as TV commercials, and whether an agency can work with it and deliver hero-quality results. The free digital twin check applies that test to a brand’s existing files.

Grip is the enterprise automation platform for publishable marketing content at scale, built by INDG. Its digital twins are the starting point for automated content: Digital Shelf generates hero shots, alternate views and promo variants from them for e-commerce, Adaptations produces compliant variations of a master asset, from Originals, a new campaign or an existing shoot, for every market and format, and Originals generates new content from a brief, grounded in the twin, with art direction and brand rules from the Brand Center applied at generation time, so the output is on-brand by design. The twins belong to the brand, and each one added makes the next launch, pack change or campaign faster to produce.

Where to start

Pick one product line and collect what you already have: CAD files, dielines, approved artwork, physical samples and the variant list. Check the gaps against the steps above, especially current artwork and a complete list of pack versions. Then twin one hero SKU and validate it against the physical product before scaling to the range. To have an existing twin assessed, use the digital twin check, or talk to the Grip team.

Frequently asked questions

How do you create a digital twin of a product?

Gather the inputs (CAD data, packaging dielines, approved artwork, physical samples and a variant list), build the geometry from CAD or scanning, define physically based materials, apply the artwork with colour management, model every variant and pack version, validate the twin against the physical product under controlled light, and store it in an open format such as OpenUSD so it can be updated when the product changes.

What are examples of digital twins?

Well-known examples include BMW’s digital twins of more than 30 production sites, Virtual Singapore’s 3D model of the city, and Dassault Systèmes’ Living Heart model of the human heart. In marketing, brands use product twins to generate packshots, configurators and campaign imagery, for example DENZA’s car configurator built from a single twin holding every variant.

What is an AI digital twin?

The term is used in several ways: AI replicas of people, such as the digital twins of models H&M announced in 2025; industrial twins used to train and test AI systems such as robots; AI models that generate 3D objects from images or text; and product twins used to ground generative AI, so that the product stays accurate while AI generates the scene around it.

How much does it cost to create a 3D product model?

It depends on the product’s shape, materials, variants and the reference material available. Published figures come mainly from vendors: Shopify put a simple 3D product model at $40 to $200 in 2020, with complex models costing more than $1,000. Transparent, reflective and highly decorated products cost more, and so do twins built to hero-campaign quality.

What file format should a product digital twin use?

OpenUSD is a common choice for production twins, because it handles complex scenes, variants and collaboration, and its core specification was published by the Alliance for OpenUSD in December 2025. glTF 2.0, an ISO standard since 2022, is widely used for web and AR viewing. A practical setup is a master twin in OpenUSD, with glTF or USDZ exports for specific channels.

‍

References

• Michael Grieves and John Vickers, “Digital Twin: Mitigating Unpredictable, Undesirable Emergent Behavior in Complex Systems,” in Transdisciplinary Perspectives on Complex Systems, Springer, 2017.
• B. Danette Allen, “Digital Twins and Living Models at NASA,” ASME keynote, 3 November 2021.
• Edward Glaessgen and David Stargel, “The Digital Twin Paradigm for Future NASA and U.S. Air Force Vehicles,” 2012.
• National Academies of Sciences, Engineering, and Medicine, Foundational Research Gaps and Future Directions for Digital Twins, 2024.
• BMW Group, “BMW Group scales Virtual Factory,” 11 June 2025.
• GovTech Singapore, “5 things to know about Virtual Singapore,” 28 March 2017.
• Dassault Systèmes, Living Heart Project.
• NVIDIA, on DENZA and WPP’s car configurator built on NVIDIA Omniverse Cloud, 8 August 2023.
• NVIDIA, Zalando and ALLSIDES case study on 3D digital twins.
• L’Oréal, “L’Oréal and NVIDIA collaborate to supercharge beauty with next-generation AI,” 11 June 2025.
• Eclat Digital, on validating materials for photorealistic product visualization, 16 January 2024.
• Karami et al., on 3D digitization of transparent objects, ISPRS Archives, 2022.
• Ben Mildenhall et al., “NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis,” ECCV 2020.
• Bernhard Kerbl et al., “3D Gaussian Splatting for Real-Time Radiance Field Rendering,” SIGGRAPH 2023.
• PlayCanvas, on relighting Gaussian splats.
• Brent Burley, “Physically Based Shading at Disney,” SIGGRAPH 2012.
• Khronos Group, glTF.
• X-Rite, Appearance Exchange Format (AxF).
• Eclat Digital, on car paint colour tolerances, 15 April 2013.
• Alliance for OpenUSD, “Alliance for OpenUSD Announces Core Specification 1.0: The Universal Language For Building 3D Worlds,” 17 December 2025.
• CG Channel, on the first OpenUSD core specification, 23 December 2025.
• Khronos Group, “glTF 2.0 released as an ISO/IEC international standard,” 2022.
• PYMNTS, on H&M’s AI digital twins of models, 27 March 2025.
• NVIDIA, on the Mega Omniverse Blueprint, 6 January 2025.
• Microsoft, TRELLIS.2.
• Meta, “SAM 3D,” 19 November 2025.
• NVIDIA, on generative AI and OpenUSD for 3D product configurators, November 2024.
• NVIDIA Developer, “Building a generative AI OpenUSD app for brand-accurate marketing visuals.”
• Shopify, on 3D models for products, 1 July 2020.