3D Modeling Services for Real-Time Web and Product Visuals
3D modeling is the production of geometry, materials and lighting that a renderer turns into an image, and in 2026 most of that output ships into a browser rather than to a render farm. Mixcore Studio builds both halves of that job: the assets themselves, and the pipeline that gets them onto a phone at an acceptable frame rate and file size.
We have shipped software for more than 8 years with a team of around 20 and 320+ delivered projects. That engineering background is the reason our 3D work is scoped in triangles, texture budgets and transfer sizes rather than in screenshots. A model that looks correct in Blender and takes eleven seconds to appear on a mid-tier Android is not a finished deliverable.
The rendering baseline moved to WebGPU
WebGPU is no longer an experiment. Chrome and Edge have shipped it since version 113 in 2023, Safari 26.0 shipped it in September 2025 across macOS Tahoe 26, iOS 26, iPadOS 26 and visionOS 26, and Firefox shipped it in 141 on Windows and extended it to Apple-silicon macOS in 145. Firefox still has no WebGPU on Linux or on Android, so Windows, macOS, iOS and Chrome-based Android are the platforms you can plan around today. WebGL 2 is now the fallback tier, not the target tier. The capability that justifies the move is compute shaders, which WebGL never had.
- Three.js r185 — the current release (npm package 0.185.x). Its WebGPURenderer attempts a WebGPU backend and falls back to WebGL 2 automatically, so one codebase covers both tiers without a separate build.
- TSL rather than raw GLSL — Three Shading Language is a node-based authoring layer that compiles to both WGSL and GLSL. New material features are TSL-first, which makes hand-written GLSL strings a migration liability if you intend to run WebGPU and WebGL from the same source.
- Babylon.js 9.0 — released March 2026 with clustered lighting, volumetric lighting, a frame graph and textured area lights. Babylon 8.0 had already moved the core engine shaders to WGSL. Note that the GLSL-to-WGSL conversion layer has not gone away: Babylon still downloads a WASM library to compile hand-written GLSL at runtime, which is why its documentation recommends authoring new shaders directly in WGSL for faster startup and smaller downloads.
- React Three Fiber v9 — the React 19 compatibility release. It accepts an async gl prop, so WebGPU initialisation wires up cleanly inside a React tree.
If an existing project of yours is WebGL-only with GLSL string shaders, it is not broken and we will not tell you to rewrite it. It is simply on the tier that is now the fallback, and any new material work in it will diverge from where the libraries are going.
Gaussian splatting became a shippable format, not a demo
Capture-based 3D crossed into standardisation this year. Khronos announced the KHR_gaussian_splatting release candidate for glTF 2.0 on 3 February 2026, with contributions from Autodesk, Cesium and Bentley Systems, Esri, Huawei, Niantic Spatial, NVIDIA and XGRIDS. Khronos targeted ratification for the second quarter of 2026, but the glTF extension registry still lists it as a release candidate, so treat it as a specification that can still change. Apple added a RealityKit GaussianSplatComponent in the 27 releases of visionOS, iOS, iPadOS and macOS, announced at WWDC in June 2026 and still marked beta in Apple's documentation.
Delivery format is where most splat projects go wrong. PlayCanvas open-sourced SOG (Spatially Ordered Gaussians), which its documentation puts at roughly 15 to 20 times smaller than an equivalent PLY, alongside Streamed SOG, a multi-LOD chunked variant for streaming large scenes, both writable from the open-source splat-transform CLI. Shipping raw PLY to a browser is a bandwidth defect rather than a stylistic choice, and it is the single most common reason a splat scene that looked good in a review call fails on a real connection.
Splats earn their place when a real object or real place has to look real and modelling it by hand is not affordable. They are the wrong tool the moment the asset must be relit, animated, configured or edited, because a splat carries baked lighting and no editable topology. A sofa that changes fabric and a room that changes time of day both need modelled geometry with PBR materials.
What we build and what we hand over
- Product models and configurators — modelled geometry with PBR materials, variant-driven colour, trim and dimension changes, and a viewer that stays under a stated transfer budget.
- Architectural and interior scenes — for web delivery, for real-time walkthroughs, or as stills where a static render is the honest answer.
- Game- and animation-ready assets — clean topology, sane UVs, LOD chains and rigs that another studio can pick up.
- Captured scenes — Gaussian splat capture, cleanup in SuperSplat, processing through the splat-transform CLI, and delivery as SOG or SPZ.
- Spatial and WebXR — WebXR is supported in Safari on visionOS, so a spatial experience can ship without Xcode or App Review, which is often the difference between a pilot happening this quarter or next year.
Interchange is glTF 2.0 for delivery and OpenUSD for authoring and exchange. AOUSD published the first OpenUSD Core Specification 1.0 on 17 December 2025, with further specification work continuing in the AOUSD Materials & Geometry and Physics working groups, which means USD is finally a written specification rather than a reference implementation you reverse-engineer. For commerce work the Khronos 3D Commerce Viewer Certification Program remains the interoperability bar for consistent material and lighting across viewers; the published certificant registry currently lists Babylon.js, Unity, Meta, UX3D, CGTrader, VNTANA and Khronos itself.
When the 3D sits inside a product interface, its accent and material colours come from the same design tokens as the rest of the UI. The DTCG Design Tokens Specification reached its first stable version, 2025.10, on 28 October 2025 and covers Display P3, Oklch and the wider CSS Color Module 4 spaces, which is what lets a rendered material and the CSS surface beside it agree on a brand colour. Worth noting for procurement language: DTCG is a W3C Community Group publication, explicitly not a W3C Standard and not on the Standards Track.
When 3D is the wrong spend
The honest position is that 3D pays back in some categories and is decoration in others, and the published evidence is uneven enough that you should insist on a pilot against your own baseline.
- Reasonably solid — the 3D Cloud Furniture Shopping Trends Study 2026, a survey of 400 US furniture purchasers, found that 78% of consumers say 3D gives them more control over purchasing decisions, and that 80% of 3D configurator users said the tool gave them confidence in their furniture-buying decision.
- Treat with caution — the conversion-uplift multiples that circulate in 3D commerce marketing are almost all vendor-published, or derived from platform data that is now several years old. We do not repeat them as forecasts, and we would treat any agency that does as selling rather than scoping.
3D is usually the wrong choice when the product photographs well and varies little, when most of your traffic is low-end Android on mobile data, when the catalogue is large and churns fast so per-SKU modelling cost never amortises, or when the goal is atmosphere on a marketing page, where a well-directed video costs less and performs better on every device. We would rather run a ten-SKU pilot measured against your own analytics than model a catalogue on the strength of someone else's case study.
Frame budget, fallbacks and accessibility
3D and motion are the two places where a demo and a real-world experience diverge most, so the deliverable includes a measured budget rather than a desktop screen recording.
- The target device is named first — typically a mid-tier Android on 4G. Triangle counts, texture resolution, draw calls, shader complexity and total transfer size are all set against that device, then measured on it.
- Documented degraded paths — what a visitor sees with no WebGPU, with no WebGL, with reduced-motion enabled, and on a slow connection. Each of those is a designed state, not an accident.
- LOD and streaming — progressive loading, level-of-detail chains and a poster frame or static render that appears before any 3D does.
- A non-3D equivalent — the European Accessibility Act has applied since 28 June 2025 across all 27 member states, and the harmonised technical bar is EN 301 549, still V3.2.1 which incorporates WCAG 2.1 Level AA. Draft V4.1.0, published in November 2025, aligns the standard with WCAG 2.2, but it is still working through the ETSI approval process and has not yet replaced V3.2.1 as the harmonised standard. In practice this means every fact the 3D view conveys, including dimensions, colour and configuration state, must also exist as text with keyboard-reachable controls.
We write the conformance target into the scope rather than leaving it implied, and for new work we target WCAG 2.2 AA directly, because writing 2.1 AA into a 2026 contract bakes in a re-audit once EN 301 549 V4.1.x lands.
What it costs to run, and who maintains it
The build figure is not the only figure. 3D carries recurring cost: CDN egress for multi-megabyte assets, asset processing and re-baking whenever the catalogue changes, and re-capture when a splat scene stops matching the real room or the real product. We quote the running number alongside the build number, because a client discovering it after launch is a failure of scoping.
The maintainability fear is well founded, and it is the question we get asked most. Commissioning a WebGPU scene with TSL shaders and a splat pipeline that only the agency can touch is a real risk. Our answer is that the handover is the pipeline, not just the files: source assets, export scripts, LOD generation, texture-compression settings, the splat-transform commands used, and the frame budget written down as a check your team can run. If your developers can regenerate an asset and verify it against the budget without calling us, the handover worked.
Our expertise
- Real-time 3D on WebGPU
- Gaussian splat capture and delivery
- Product models and configurators
- PBR texturing and material accuracy
- glTF and OpenUSD asset pipelines
- Frame budgets and LOD
Frequently asked questions
Should we build on WebGPU or WebGL in 2026?
Build on WebGPU with WebGL 2 as the fallback. Chrome and Edge have supported it since version 113, Safari 26.0 shipped it in September 2025, and Firefox shipped it in 141 on Windows and 145 on Apple-silicon macOS. Firefox still has no WebGPU on Linux or Android, which is exactly why the WebGL 2 fallback stays in scope. Three.js WebGPURenderer falls back to WebGL 2 automatically, so a single codebase serves both tiers. The reason to prefer WebGPU is compute shaders, which WebGL never offered.
What is Gaussian splatting, and when should we use it instead of a modelled asset?
Gaussian splatting renders a scene as millions of small oriented blobs captured from photographs, rather than as polygons. Use it when a real place or object has to look real and hand-modelling it is not affordable. Avoid it when the asset must be relit, animated, edited or configured, because a splat has baked lighting and no editable topology. Khronos published the KHR_gaussian_splatting release candidate for glTF 2.0 on 3 February 2026, so it is now heading for a standard rather than being a per-vendor format.
Will a 3D product viewer actually work on a mid-range Android phone?
Only if it was budgeted for one from the start. We name the target device before modelling begins and set triangle counts, texture resolution, draw calls and total transfer size against it, then measure on the real device rather than on a desktop. We also design the degraded states explicitly, covering no-WebGPU, no-WebGL, reduced-motion and slow connections. A viewer without a documented fallback is a demo, not a shipped feature.
Does 3D actually increase conversion?
It depends on the category, and the published evidence is uneven. The 3D Cloud Furniture Shopping Trends Study 2026, a survey of 400 US furniture purchasers, found that 78% of consumers say 3D gives them more control over purchasing decisions and that 80% of 3D configurator users said the tool gave them confidence in their furniture-buying decision. The conversion-uplift multiples widely quoted elsewhere are mostly vendor-published or based on platform data that is now several years old, so we do not repeat them as forecasts. We recommend a small pilot measured against your own baseline before committing a catalogue.
Can our in-house team maintain the 3D assets after handover?
That is the design goal. The handover is the pipeline rather than just the files, so you receive source assets, export scripts, LOD generation, texture-compression settings, any splat-transform commands used, and the frame budget written down as a check your team can run. We also quote the running cost, including CDN egress, asset reprocessing and re-capture, so the cost of ownership is known before launch rather than discovered after it.
Contacts
We are always happy to talk with you.
Feel free to contact us in any suitable way
Request a quote
Let's discuss your project!
Please, provide us with a brief description of what you
already have and what you are going to achieve.
Mail us contact@brainiacminds.com