(Notes)

A Guide to 3D Web Design: Cost Drivers, Process and Technology Choices

The outcome of a 3D web design project depends on model source, interaction depth, and the supported phone range. These three set the cost and also shape the technology and the acceptance method.

Polish Design engineering team・2026-11-16

Contents

3D web design places a product, a space, or an event in the browser so that visitors can rotate, zoom, and switch states with a mouse or a finger. The workload is set by three things: where the models come from, how deep the interaction goes, and how far back the supported phones reach. Those same three things set the cost, the technology choices, and the way the work is accepted. Each subtopic has its own article, linked in the text, so the details are not repeated here.

Reading order: the sections on what the work includes and the three common types help identify which kind of project applies. Technology choices and process follow. Model acceptance, cost, performance, and search each get a short summary, with details in the matching article. A checklist for contract review closes the guide. For a direct scope assessment, see the description of 3D web design services.

What 3D web design includes

A 3D website involves six kinds of work, each with a different responsible party.

Work typeContentPrimary responsible party
Models3D files of the product or scene, sourced from a rebuild by the modeler, a CAD conversion, or a scanDelivered by the modeler; intake and acceptance by the development team
SceneLighting, materials, camera, loading order, and the logic for swapping colors and partsDevelopment team
InteractionRotate and zoom, configuration, exploded views, section cuts and floor selection, guided toursDevelopment team; content confirmed by the brand
Web pages and searchText in HTML, titles and descriptions, share previews, structured dataDevelopment team; copy supplied by the brand
PerformanceFirst-screen download size, memory, per-frame time, and device tiersDevelopment team; device range set by the brand
Admin and maintenanceContent updates, model replacement, monitoring and fixesBrand and development team, divided by contract
Fig. 1The brand sets scope, the modeler supplies raw material, and the development team handles intake, acceptance, and launch
Requirements and assets
Model intake
Acceptance and launch
Brand
Modeler
Development team
1
BrandDevice range and content ownerDecides how old a supported phone can be
2
BrandExisting models or drawingsRebuilt by the modeler when none exist
3
ModelerDelivered models and source files
4
Development teamItem-by-item comparison and intake acceptanceMissing faces, duplicate exports, extra nodes
5
Development teamAcceptance record for measurable items
6
BrandConfirm trade-offs and sign off

One typical arrangement of the three parties; the actual division follows the contract.

Three common types

Two people stand on either side of a conveyor belt carrying several containers with yellow lids, each reaching to handle one.
  • Product display and configuration: visitors rotate the product and swap colors and parts. For a folding bicycle configurator, the approach was to map the 52-step CAD exploded sequence supplied by the client onto a 31-segment web timeline, and to list what could not be done before work began (3D configurator case).
  • Space and property tours: exterior models split into bundles, floor selection, and interior panoramas, with smaller textures and lighter versions on phones. See how a 3D property showroom site is built. For a 3D city homepage whose phone memory budget was converted line by line, see the 3D city homepage case.
  • Exhibitions and events: large volumes of content that need navigation and search. The online expo drew a 3D world map in Three.js and handled search inside the browser (3D world map for an online expo).

Choosing the technology

WebGL is the JavaScript graphics API built into browsers. It needs no plug-in, draws 2D and 3D graphics inside a canvas element, and uses hardware acceleration when the device supports it (MDN Web Docs, 2026). Three.js is a widely used JavaScript 3D library, and glTF is a transmission format for 3D assets; section 2.3 of the specification calls glTF open and interoperable, and easy to process and render at runtime (Khronos, 2021). WebGPU is a newer browser graphics API. The role of each term and how the terms differ are covered in how WebGL, WebGPU, and Three.js divide the work.

At an exhibition, visitors hold a 3D product on a phone next to the real one.At an exhibition, visitors hold a 3D product on a phone next to the real one.
At an exhibition, visitors hold a 3D product on a phone next to the real one.

Whether a custom 3D scene is needed depends on what visitors have to do. When no interaction is needed, a video or a still image loads faster and uses no phone memory. When the visitor only needs to rotate a single model, an off-the-shelf model viewer is enough and takes the least development. A custom scene is warranted for swapping colors and parts, section cuts, animation timelines, or links to the rest of the page.

Fig. 2Only content that needs interaction goes into a custom 3D scene; the rest uses video, images, or a model viewer
  1. Confirm what visitors need to doView, rotate, or configure and interact
  2. Confirm target devicesHow old a device must be supported
  • No interaction needed
    Video or still imageFastest to load, no memory use
  • Rotating a single model is enough
    Model viewerLeast development
  • Swapping colors and parts, section cuts, timelines, or page-linked behavior
    Custom 3D sceneThree.js on top of WebGL, assets in glTF

Lists decision criteria only and does not compare products.

Process and the output of each stage

The process has six stages. Each stage produces a defined output, and the output of one stage is the input of the next.

A team member pushes hard on the needle of a large gauge.
Fig. 3Six gates passed in sequence, each handing over an output that can be checked
  1. Requirements and device rangeTarget device list, what is being shown, content update owner
  2. Model intake and acceptanceIntake comparison record, deployable files reproducible from source
  3. First-screen budgetLine-by-line budget for first-screen download size and memory
  4. Interaction and animationInteraction list, list of items that cannot be done
  5. Search and share previewText in the source, title, description, share preview
  6. Launch and maintenanceAcceptance record, procedure for updates and model replacement

The type of output at each stage.

Model intake and acceptance

Model files from the modeler are raw material. The deployable files need to be reproducible from the source, and intake compares missing faces, duplicate exports, and extra nodes item by item (intake, reproduction, and acceptance of outsourced 3D models). The reasoning is that 3D failures mostly raise no error; a part is simply missing. Every acceptance item therefore needs measurable evidence, and checking a screenshot by eye does not count. When a file is too large, textures come first and geometry second; the order is in handling an oversized GLB file.

An engineer checks a model part by part against an acceptance checklist.An engineer checks a model part by part against an acceptance checklist.
An engineer checks a model part by part against an acceptance checklist.

What determines cost

Cost differences fall in three places: whether the models must be rebuilt, how deep the interaction goes, and how far back phone support reaches. For models with complete sources that pass intake, the main work is optimization and launch; for models with unclear sources, repair is listed as a separate item first. The deeper the interaction, the more scene logic and acceptance items there are. The brand decides the phone support range, and each additional older device adds another round of memory budget conversion.

Another place that shapes the total on a quote is the scope clause. Fixes for functions that break within the contract and new requests beyond the scope should be written in two separate columns before signing. The breakdown of the three drivers and how to read a quote are in what determines the cost of a 3D website; this guide lists no amounts. To learn the service scope first, see 3D website development services.

How performance is accepted

Performance is measured in three parts: loading, memory, and per-frame time. The measurement methods and the fixes do not carry over from one part to another (Web 3D performance engineering). For loading, the good threshold for LCP is 2.5 seconds or less (Walton et al., 2019). The LCP definition lists images, text blocks, and video, however, so the moment a product first appears inside a canvas needs a separate measurement point. For interaction, INP is good at 200 milliseconds or less and poor at 500 milliseconds or more (Wagner et al., 2022).

The MDN WebGL guidance has three parts: reduce draw calls, budget texture memory, and compile and link shaders in parallel (MDN Web Docs, 2026). Draw calls and shader compilation land in the frame loop, while textures land in memory and loading. The documentation gives principles and no measured numbers, so the numbers have to be measured on the project.

First-screen download size is inventoried in rounds with a fixed method. For a folding bicycle configurator, after the build, a headless browser read every resource request and grouped the requests by file extension; 3D files made up about 80%. Each round touched only data not needed at runtime, and the first-screen download dropped from about 57 MB to about 43 MB, then to about 37 MB.

The memory budget is set by the client at the start of the project and converted line by line. For a property showroom site under Chromium mobile emulation, memory across the full flow dropped from about 2.6 GB to about 0.9 GB, and the client then asked that the peak of the flow stay under 600 MB (memory budgets for mobile 3D). These numbers come from emulation, not from physical devices. The method and error margins for per-frame time are in measurement error and reliable methods for 3D performance numbers.

Search and AI crawlers

For a front-end-only 3D website, a crawler often receives an empty HTML shell at first. Google processes JavaScript pages in three phases (crawling, rendering, and indexing) and states that server-side rendering and pre-rendering remain worthwhile, partly because not all bots can run JavaScript (Google Search Central). The deciding factor is that text content must appear in the source; the 3D view itself does not earn search ranking.

Pre-rendering and crawler routing are covered in search and AI crawler readability for 3D websites. The differences among SEO, AEO, and GEO and the order of implementation are in what SEO, AEO, and GEO are.

Checklist before signing

  • Model source: whether the deployable files can be reproduced from the source files, and who keeps the source files, should be stated in the contract.
  • Acceptance items: whether each item has measurable evidence and stated measurement conditions, including target devices, network throttling, cache state, and sample count.
  • Phone device range: how old a supported device can be, and who sets and measures the memory budget.
  • Maintenance scope: whether fixes for broken in-contract functions and new requests are in separate columns, and how each is priced.
  • Contradictions between clauses: for example, one clause states that the listed items are the complete delivery scope while another says "including but not limited to"; the two need to be aligned before signing.

When 3D is not a fit

  • The product needs only three to five good photos, with no decision that depends on rotation or customization.
  • The main users are on low-end phones, and tiered presentation is not acceptable.
  • The content changes heavily and often, with no one to maintain models and assets.
  • The expectation is that 3D alone will directly earn search ranking.

What to prepare

  • What is being shown: a product, a space, or an event.
  • Existing models or drawings.
  • Target devices and traffic sources.
  • The number of options to compare.
  • Who owns content updates.

Once the material is ready, ask about a 3D web design assessment.

(Contact)

Start a project.

Three steps to a clear direction and budget range.

  1. 01Book the first meeting: choose a time. A full specification is not required.
  2. 02A 15 to 30 minute online consultation: direction, budget range, and draft timeline.
  3. 03After materials are reviewed, a quote and delivery specification follows in about 3 to 5 working days.
Email
hello@polish-design.com.tw
Mailing address
3F, No. 191-1, Zhifu Rd., Zhongshan Dist., Taipei 104, Taiwan

Book the first meeting

Monday to Friday, 16:00–18:0030 minutes online

Choose a time, then confirm with a Google account or an email address. A confirmed booking can be added to the calendar.

Choose a time

Or send the details first

Not ready to pick a time yet: leave the project details here and the team replies by email. A meeting can still be booked right after sending.

More details (optional): service, stage, links, source

Information is used only to evaluate this project.

ISO/IEC 27001:2022 certificate

(Certificate)

ISO/IEC 27001:2022

Certification body
ARES International Certification
Scope
The development process for interactive immersive 3D systems, and network and infrastructure operations and management.

Verify with ARES

Urbird

Urban renewal you can see.

Before construction begins, landowners can walk through their new homes. From consent forms to demolition notices, everything runs in one system.

More than 100 modules, AI features included, combined to each developer’s needs.