Amazon 3D model requirements, Shopify’s upload rules, Google’s merchant specs — this is the reference we wished existed every time a client asked “will our model actually work on our store?” The platforms publish their rules in scattered docs, many public summaries repeat at least one outdated or second-hand requirement, and guessing can cost you a rejected upload plus another review cycle of up to two weeks. Everything below was checked against the platforms’ own documentation, including Amazon’s own specification read inside Seller Central on 5 August 2026 — which is where the numbers repeated by many public guides turn out to be wrong.
Shopify 3D model requirements
Shopify accepts GLB and USDZ as product media, with a 500 MB upload ceiling — but the number that actually matters is 15 MB: every uploaded model is converted so that both a GLB and a USDZ exist, but anything above 15 MB is also automatically optimised — Shopify’s wording is that “model geometry and textures are optimized, while preserving the visual output of the original file”. Conversion happens either way; the optimisation is the part you don’t control, which is why professional delivery stays well under the threshold. Shopify’s own artist checklist targets roughly 4 MB, with textures at 2048×2048 or below for mobile, and requires PBR maps packed in the ORM convention (occlusion, roughness, metalness in one texture’s channels). Two conveniences worth knowing: upload one well-made GLB and Shopify generates the USDZ itself, covering the 3D viewer, Android AR, and Apple Quick Look from a single file — Shopify’s storefront model viewer uses Google’s <model-viewer> component — and the same artist checklist makes GLB the standard handoff, with a separate USDZ produced only when a merchant or an older workflow specifically asks for one.
Amazon 3D model requirements — and the error every blog makes
Start with the correction, because most published guides get it wrong in the first paragraph: Amazon accepts GLB and glTF only. USDZ is not on the menu. That’s per Amazon’s own seller-forum guidance from its community management, which also lays out the path: uploads go through Seller Central’s Image Manager (Catalog → Upload Images → 3D models), the account must hold the appropriate Amazon Brand Registry role for that product — registered brand owner or Brand Representative, and most submissions are reviewed within about two weeks. Eligible categories center on home, furniture, consumer electronics, shoes, and eyewear — the last two feeding Virtual Try-On.
Now the part worth the price of admission. Amazon’s technical spec sits behind a Seller Central login, which is why almost every public guide repeats the same second-hand numbers — 20,000 triangles, 1024×1024 textures. We read the actual document. Both are wrong, and one of them will get your model rejected.
Amazon’s published requirements, from the Seller Central 3D guidelines:
- Triangles: up to 200K — ten times the circulated figure. The budget is far more generous than the internet believes.
- Textures: 2K minimum, 4K maximum — “not be less than 2K (2048 px x 2048 px), and no larger than 4K (4096 px x 4096 px)”. So the widely repeated 1024×1024 is below Amazon’s floor. Build to the circulated spec and you fail the one you were trying to satisfy. Textures must be square, power of two, one per channel, and double-sided textures are not supported.
- Formats: .GLB and .glTF, with GLB the recommended one. Exported glTF must ship as separate files (.gltf + .bin + textures) — Amazon does not accept glTF with image resources embedded in the JSON.
- Materials: PBR metal-rough throughout. BaseColor, Metallic and Roughness maps are required; Normal, Opacity, Height, Thickness and Occlusion are optional. UVs in a single atlas within (0,1), maximised.
- Clean file: no modifiers, no animation keyframes, no cameras, no lights, and no invalid triangles, flipped normals or stray vertices. Real-world scale, matching assembled product dimensions — Amazon warns that inaccurate dimensions delay publication.
- Alignment matters more than people expect: up is +Y, model front faces +Z, pivot at 0,0,0, and the mesh must be aligned to its product type — floor, wall, ceiling or floor-mirror. Get this wrong and publication is delayed or blocked outright.
- Seven glTF extensions are listed as supported:
KHR_materials_sheen,KHR_materials_transmission,KHR_materials_clearcoat,KHR_materials_ior,KHR_materials_volume,KHR_draco_mesh_compressionandKHR_interactivity, with the note that unlisted extensions are not currently supported — they may block publication or display incorrectly. Older public Amazon forum guidance names a different and longer set, so treat the Seller Central document for your own account and category as the controlling source. - No human imagery anywhere in the asset.
- Amazon points at the Khronos glTF-Validator to check files before submission — its export page words this as a requirement, its tools page as a suggestion. Run it either way.
We could not find a stated file-size cap in the general guidelines, which is the one number the second-hand guides agree on (5 MB) and the one we cannot confirm. Treat that as unverified; everything above is Amazon’s own wording. Category-specific guides carry extra rules on top — eyewear, for instance, requires named glTF nodes for frame, temples and lenses.
Google, Meta, and everyone else
Google Merchant Center takes 3D models via the virtual_model_link attribute — note the name; a surprising share of guides cite an attribute that doesn’t exist — in .gltf or .glb, recommended under 10 MB with a 15 MB maximum, and the model must sit at a crawlable URL. Google suggests rather than requires the glTF Validator — “you can use a validation tool to verify” is the actual wording. Textures are where Google contradicts itself: the Merchant Center page recommends 4K, while the Scene Viewer file requirements it links to cap textures at 2048×2048. Build to the lower number and you satisfy both. Review takes up to two weeks, and coverage is limited to home goods and shoes. Geography is Google’s second self-contradiction: the attribute specification says US only, while Google’s own 3D-and-AR overview says “the US, CA, AU, JP and IN”. Check eligibility inside the target Merchant Center account before promising anyone a deployment date. Meta is the cautionary tale: Meta Spark shut down in January 2025, so any guide still describing how to publish merchant AR effects through Spark is describing a route that no longer exists. WooCommerce and BigCommerce document no Shopify-style native 3D product media in their standard product tools — implementations there run through extensions, marketplace apps, configurator platforms or custom model-viewer embeds, which at least means the same GLB discipline applies.
One master asset, three platform exports
Put them side by side — Shopify’s ~4 MB artist target, Google’s 10–15 MB ceiling, and Amazon’s 200K triangles with 2K–4K textures — and the strategy writes itself: build one approved source asset to the strictest requirement on each axis, then export per platform. It is not one GLB submitted everywhere: Shopify wants ORM-packed textures while Amazon asks for separate greyscale metallic and roughness maps, and Amazon adds its own axis, pivot, alignment and extension rules on top. The goal is to avoid rebuilding the product, not to force one identical file through three different gates. Note the axes differ. Shopify and Google constrain file size; Amazon constrains geometry and texture resolution, with a floor as well as a ceiling. Since Meta’s exit, GLB has become the dominant shared commerce format — with USDZ still needed for some Apple workflows, and generated automatically by Shopify from whatever you upload — and there’s now an industry-standard playbook for making assets that survive all of it: the Khronos Group’s Asset Creation Guidelines 2.0 for commerce-ready glTF assets, published August 2025. Build to the loosest gate instead, and every new channel is a rebuild.
What actually gets a file through review
The review gates — Amazon’s two weeks, Google’s two weeks — are where timelines die, because a rejection adds another correction-and-review cycle to the schedule. Technical compliance improves your odds; it does not guarantee a pass, since both platforms also review quality, eligibility, brand ownership and category rules. The pre-flight list that removes the most common avoidable failures:
- Real-world scale — a sofa that imports at four metres tall will fail validation or hold up publication.
- A clean pass through the glTF Validator.
- Texture packing and resolution matched to the destination platform, not to a house default.
- No stray cameras, lights, or animations left in the file.
- A visual check in an actual viewer, not a DCC viewport.
This is the unglamorous half of product visualization work — the same models that carry your product rendering only earn their keep on-platform when the delivery files respect the gates above. Specs in this article were checked against current platform documentation between 31 July and 5 August 2026, including Amazon’s authenticated Seller Central guideline on 5 August; platforms revise these, which is half the reason spec-compliant delivery is a service and not a checkbox.