3D Product Models for E-commerce: Cost & Specs

The short version: Commissioning 3D product models for e-commerce is a procurement problem, not a technical one. There are three production routes — photogrammetry, manual modelling from CAD or drawings, and AI image-to-3D — and they differ enormously in accuracy, turnaround and how much cleanup lands back on you. We are not going to publish a price table, because the honest range is too wide for one to be anything but misleading. What we will give you is the list of things that actually move the price, a specification you can paste straight into a brief, and a checklist for rejecting a model before it reaches your product pages.

Why nobody publishes a price list

Search for what 3D product models cost and you will notice something: almost every result is a company that makes them, and almost none states a price. That is not evasion for its own sake. It is because the same phrase — "a 3D model of a product" — covers work that genuinely differs by an order of magnitude.

A plain cylindrical candle in a single colourway and a motorcycle helmet with visor mechanism, three material finishes and a printed graphic wrap are both "one product model". One is close to trivial. The other is a week of skilled work. Any single published figure would be wrong for most readers, and the ones you do occasionally see quoted tend to be either a loss-leader entry price or an enterprise average, neither of which describes your job.

So rather than invent a number, here is the thing that is genuinely transferable: the variables. If you understand what makes a quote move, you can read any quote you are given and tell whether it is sane.

The three ways a model gets made

Pick the wrong method for your product category and you will pay twice — once for the model, and again for the rework when it does not hold up on a product page.

Photograph the real thing, reconstruct it

Dozens or hundreds of photographs of a physical sample from every angle, processed into a mesh and texture. The output inherits the real object's surface detail, which is why it is unbeatable for anything irregular or organic.

Strongest for: textiles, footwear, food, natural materials, anything hand-made or irregular
Weakest for: glass, chrome, mirrors and anything highly reflective — the maths struggles badly with surfaces that do not look the same from two angles
You must provide: a physical sample, shipped and returned
Hidden cost: retopology. Raw scan output is a dense, messy mesh and almost always needs rebuilding before it is fit for the web

A practical hybrid most people miss: use AI generation to produce rough models for the whole catalogue quickly, put them in front of your team, and identify the twenty SKUs where 3D actually changes the buying decision. Commission those properly. You will spend a fraction of what a full catalogue would cost and end up with better assets where they matter.

What actually moves the price

When a quote comes back higher than you expected, it is nearly always one of these.

Geometric complexity

Moving parts, internal structure, perforations and fine mechanisms all multiply the modelling time. A closed box is not a hinged case.

Material count

Each distinct material needs its own set of texture maps. A product in one matte plastic is far cheaper than one combining brushed metal, leather, stitching and translucent plastic.

Reference quality

Good CAD or dimensioned drawings can halve the work. A handful of phone photographs and a guess at the dimensions will do the opposite.

Accuracy requirement

"Looks right on a product page" and "dimensionally correct for AR placement in a customer's room" are different jobs with different price tags. Say which you need.

Variants and configurability

Cheap if planned into the base model, expensive if bolted on afterwards. Decide before the build starts, not after.

Revision rounds

Often the single largest overrun. Unlimited revisions are not free — they are priced into the quote, or they become an argument later.

Getting quotes you can compare

The most common procurement failure here is collecting three quotes that are not for the same thing, then choosing the cheapest and discovering why it was cheapest.

Quote one representative product, not the catalogue. Pick something typical rather than your simplest item. Per-unit price on a hundred models is meaningless if the sample was your easiest SKU.

Send an identical brief to everyone. The specification below exists for this. Differences in quotes should reflect the supplier, not differences in what you asked for.

Ask what happens on revision three. Get the number of included rounds and the cost of additional ones in writing. This is where budgets quietly die.

Ask who owns the source files. Delivery of a GLB is not the same as delivery of the working file. Without the source you cannot make changes later without paying to rebuild — and you may not be able to change supplier at all.

Ask for one paid sample before the bulk order. Any competent supplier will agree. Paying for one model to de-risk a hundred is the cheapest insurance in this whole process.

The question most buyers forget: who owns the copyright and the source files? Get it stated explicitly. We have seen retailers discover, two years and several hundred models in, that they licensed rather than bought their entire 3D catalogue and could not move supplier without starting again.

The specification to send

Copy this, fill in the bracketed parts, and send it to every supplier you approach. It removes most of the ambiguity that causes disputes later.

Deliverables — GLB (primary), plus source file in native format — USDZ if the target platform does not generate it — All texture maps as separate files Technical constraints — Target file size: 4 MB or under per model — Texture maps: no larger than 2048 x 2048 px — PBR workflow, textures embedded in the delivered GLB — Quad-based topology unless the format requires triangles — Watertight, manifold geometry. No open edges, no duplicate vertices — Model must be correctly scaled in metres and origin-centred at the base Accuracy — Dimensional tolerance: within [X] mm of physical product — Colour to match [Pantone / supplied sample / brand guideline ref] Variants — Base model must support material swaps without re-modelling — Colourways required: [list] Commercial — Revision rounds included: [N]. Cost per additional round: [state] — Full IP and source files transfer to us on final payment — Delivery deadline: [date]. One paid sample before bulk order

The two lines that save the most money are the scale requirement and the source-file transfer. Wrong-scale models are the most common defect in AR retail, and not owning your source files is the most expensive thing to discover late.

Platform requirements

Design to the tightest target you actually need to hit. For most UK retailers that is Shopify's guidance, which is unusually specific and worth quoting exactly rather than paraphrasing.

Requirement	What Shopify states	Design to
Accepted formats	GLB and USDZ	Deliver GLB; Shopify generates the other
Maximum file size	Up to 500 MB	Ignore — a limit you should never approach
Auto-optimisation	Applied above 15 MB	Stay well below, so you control the quality
Target file size	Around 4 MB	This is the real budget. Write it into the brief
Texture size	No larger than 2048 x 2048 px for mobile web	2048 max; drop to 1024 where detail allows
Topology	Quads preferred unless format requires triangles	Quads, for cleaner edits later
Polygon count	No specific figure published	Judge by file size and device performance instead

On polygon limits. A great deal of published guidance states that Shopify enforces a polygon ceiling, usually quoted as 100,000. We could not locate that figure in Shopify's documentation. Its 3D modelling guidance advises keeping geometry as low as possible while retaining necessary detail and respecting the polycount limits of the target format, without naming a number. Treat any specific figure you are quoted as a supplier's working rule, and hold them to file size and real-device performance instead — those are the things a shopper actually experiences.

4MB target file size per model

2048px maximum texture dimension

1GLB you need to deliver

3production methods to choose between

How to reject a bad model

Run every delivered model through this before it goes anywhere near a live product page. It takes about ten minutes and catches nearly everything.

Measure it in AR. Place it on a real floor next to the real product or a known reference. Wrong scale is the most common and most damaging fault, and it is invisible on a desktop screen.

Check the file size against the brief. If you asked for 4 MB and received 30 MB, it is not finished, regardless of how good it looks.

Rotate it fully and look underneath. Missing faces, inverted normals and untextured undersides usually hide on the side the reference photographs did not cover.

Look at it in two lighting environments. A model tuned to flatter one studio light setup can look plastic or muddy in another. AR lighting comes from the customer's actual room.

Compare colour to the physical product in daylight. Not on your monitor alone. Colour drift is the fault customers notice after delivery, which is exactly the return you were trying to prevent.

Load it on a mid-range phone on a throttled connection. Not your laptop. If it stutters or takes several seconds, it is too heavy for the job.

Procurement mistakes that cost real money

Modelling the whole catalogue

Start with the SKUs where a shopper genuinely cannot judge the product from photographs. Most catalogues have far fewer of those than the initial enthusiasm suggests.

Not specifying scale units

Centimetres and metres confusion is endemic. State the unit and the tolerance in the brief, then verify it on delivery.

Buying delivery, not ownership

Without source files and IP transfer, every future change is a rebuild and switching supplier is close to impossible.

Variants as an afterthought

Six colourways designed in are a materials change. Six colourways bolted on afterwards are six models.

No named approver

Sign-off by committee produces revision rounds nobody budgeted for. One person decides, or the quote is fiction.

Skipping the paid sample

Committing to a hundred models on the strength of a portfolio rather than one job done for you is how bulk orders go wrong.

Frequently asked questions

How much does a 3D product model cost?

There is no useful single figure, which is why almost no supplier publishes one. The same phrase covers a simple single-material object and a complex assembly with moving parts and multiple finishes, and those differ by an order of magnitude in effort. The variables that move the price are geometric complexity, the number of distinct materials, the quality of the reference data you can supply, whether you need dimensional accuracy for AR or only visual plausibility, how variants are handled, and how many revision rounds are included. Get quotes on one representative product using an identical brief, and the comparison becomes meaningful.

What file format do I need for 3D products on a website?

GLB is the primary web format — a single binary file containing geometry, materials and textures. USDZ is Apple's format, needed for AR Quick Look on iPhone and iPad. Shopify accepts both and generates whichever you did not supply, so a single GLB is usually enough. Always ask for the source file in its native format as well, because without it any future change means rebuilding from scratch.

Is AI 3D model generation good enough for e-commerce?

It is genuinely useful for volume and for prototyping, and it is improving quickly, but its weakness is dimensional accuracy. Output often looks convincing while measuring incorrectly, which is acceptable for a spinning viewer and unacceptable for AR placement at real scale, where a wrong-sized product actively misleads the customer. A sensible approach is to use AI generation to produce rough models across the catalogue, identify the SKUs where 3D genuinely changes the buying decision, and commission those properly.

What size should my 3D model files be?

Aim for around 4 MB per model, which is the target Shopify's own guidance states, with textures no larger than 2048 by 2048 pixels for mobile web. Shopify accepts files up to 500 MB and automatically optimises anything over 15 MB, but relying on that means the platform makes quality decisions for you. Texture data usually dominates file size rather than geometry, so halving texture resolution is normally the fastest way to bring a heavy model within budget without visible loss on a phone screen.

Do I need to send a physical sample?

Only for photogrammetry, which reconstructs the model from photographs of the real object and therefore requires it. Manual modelling works from CAD files, technical drawings and reference photography, so if your product was designed digitally the source data already exists and no sample is needed. Budget for shipping in both directions when photogrammetry is the right method, and confirm who is responsible for the item while it is with the supplier.

Who owns the 3D models once they are made?

Whatever the contract says, which is why it needs stating explicitly rather than assuming. Some suppliers deliver a GLB while retaining the source files and the underlying IP, meaning you cannot make changes or move supplier without commissioning the work again. Specify full transfer of copyright and source files on final payment in your brief, and confirm it in writing before the first model is produced.

Want a sanity check on a quote you have been given?

Send us the brief and the number. We will tell you whether it looks reasonable for the work described, and if the honest answer is that you do not need 3D models for most of your catalogue, we will tell you that instead.

Ask us to look at it

3D Product Models for E-commerce: Costs, Specs and How to Brief a Supplier