A metaverse 3D space is an interactive environment, such as a showroom, venue, campus, store or replica of a real site, that people explore in a browser or headset. Building one is mostly an art and optimization problem: creating or capturing the environment, then reducing it until it loads in seconds and runs smoothly on the weakest device you support.
This page covers the environment itself. For multiplayer, voice and backend systems, see metaverse application development; for avatars, see metaverse avatar development.
Three ways to create a space
| Method | How it works | Best for | Watch out for |
|---|---|---|---|
| Modeled by artists | Built in Blender, Maya or 3ds Max, or converted from CAD/BIM | Stylized or idealized spaces, planned buildings, anything interactive | Artist time; CAD files are far too detailed and need heavy cleanup |
| Photogrammetry or lidar scan | Many photos or a scanner produce a textured mesh of a real place | Replicas of real sites, heritage, real estate | Messy meshes, baked-in lighting, large textures |
| Gaussian splatting | Photos or video are turned into millions of soft 3D points that render photorealistically | Photoreal viewing of real places and objects | No clean collision or editing; still maturing in engines and web renderers |
Hybrids are common: a splat or scan for the photoreal backdrop, with modeled geometry for floors, walls and anything users interact with.
The optimization pass
The difference between a space people enjoy and one they abandon is usually load time and frame rate. A practical checklist:
- Geometry: remove hidden faces, decimate dense meshes, merge static objects, and create levels of detail for large items.
- Textures: pack into atlases, cap resolution by how close users can get, and compress with KTX2/Basis Universal so they stay compressed in GPU memory.
- Lighting: bake lighting and ambient occlusion into lightmaps instead of computing dynamic lights at runtime. Use a small number of real-time lights only where things move.
- Draw calls: fewer materials and merged meshes mean fewer draw calls, which matters most on mobile and standalone headsets.
- Compression: Draco or meshopt for geometry in glTF files.
- Streaming: load the entrance area first and stream other rooms in as visitors approach.
- Collision: use simple invisible collision shapes rather than the visual mesh.
Formats
Use glTF 2.0 (as GLB) for real-time delivery to web and most engines. Use OpenUSD when the space is part of a larger production or industrial pipeline, especially with NVIDIA Omniverse or Apple platforms. Keep the editable source files (Blender, Maya, USD layers) under version control; exported files alone are hard to maintain.
Delivering the space
Browser
Three.js, Babylon.js and PlayCanvas all load glTF and support WebGPU or WebGL. A browser space can be shared as a link, embedded in a website and opened in a headset via WebXR. Target a first interactive view within a few seconds on a mid-range phone, and show something meaningful while the rest loads.
Native or streamed
Unity or Unreal builds give more visual headroom and are the norm for VR. Unreal's Pixel Streaming can deliver a high-end render to any browser, at the cost of running a cloud GPU for every concurrent user.
Making the space useful
A beautiful empty room is a screensaver. Decide what visitors should do: inspect products (clickable hotspots, configurators, 3D product models), attend something (screens with live streams, stages), meet people (spatial voice zones), or learn something (guided tours, information panels). Hook these to a CMS so content changes do not require a new build, and add analytics to see which areas people visit and where they leave.
Process and effort
- Define purpose, target devices and performance budgets.
- Block out the layout with simple shapes and test navigation.
- Create or capture the environment.
- Optimize, bake lighting and export.
- Build interactions and connect the CMS.
- Test on real devices and iterate.
A single polished room or showroom typically takes a 3D artist and a developer around four to eight weeks; a multi-area campus or detailed replica of a real site can take several months. Spaces built for specific uses are covered in the guides to metaverse art galleries and virtual event platforms.
Does a 3D space need a blockchain? Almost never. The exception is when visitors own and trade items placed in the space, or the space itself is a tokenized parcel in a world like those discussed under Decentraland-style platforms.
Frequently asked questions
Can I turn my CAD or BIM model into a metaverse space?
Yes, but expect significant cleanup. Engineering models contain far more detail than real-time rendering can handle, as well as hidden internal parts. They need decimation, material simplification and lighting work.
Are Gaussian splats ready for production?
For photoreal viewing, often yes, and support in web and engine renderers keeps improving. For interactive spaces, combine them with conventional meshes for collision and interaction.
How big can a browser-based 3D space be?
There is no hard limit if you stream areas progressively, but the initial download should stay small enough to load within a few seconds on a phone. Plan around that budget.
Should the space be built for VR?
Build it so it works on screens first, then make sure it is comfortable in VR: real-world scale, teleport or smooth movement options, and a sustained high frame rate.