The short answer: 3D Gaussian Splatting (3DGS) turns a capture of a real space into a photorealistic 3D scene that renders smoothly in an ordinary browser. In Singapore you can produce 3DGS scenes by renting XGRIDS equipment: the PortalCam at S$1,200/week for purely visual capture, or the Lixel K2 at S$2,300/week when you also need measurable point cloud data, and processing through the LCC CyberColor workflow.
What is 3D Gaussian splatting?
Instead of triangles and textures, a 3DGS scene is millions of tiny oriented, semi-transparent Gaussian splats, each carrying position, shape, colour, and opacity. A training process optimises them against the captured imagery until, rendered together, they reproduce the space with the soft realism of photographs, reflections, foliage, and fine detail included, while remaining fully navigable in 3D. Because splats rasterise the way ordinary graphics do, the scene runs in real time in a browser tab on normal hardware, no special viewer or plug-in required. It is the first reality-capture format that looks right to non-technical stakeholders and is trivial to share: a link, not an installation. That combination, photographic realism plus frictionless delivery, is why 3DGS moved from a 2023 research paper to production pipelines faster than almost any capture format before it.
3DGS vs NeRF: what changed?
Neural radiance fields (NeRFs) proved that photoreal novel views could be reconstructed from ordinary photos, but the scene lived inside a neural network that had to be queried view by view, so rendering was slow and hosting was awkward. 3DGS, introduced in 2023, keeps the photorealism but swaps the representation: explicit splats instead of network weights. The practical consequences are the ones that matter commercially: real-time rendering on ordinary GPUs, straightforward web delivery, and editable scenes. For teams choosing today, the debate is largely settled: NeRF remains a research lineage, 3DGS is the deployable format. How both compare to classic photogrammetry is covered in photogrammetry vs LiDAR vs 3DGS.
What do you capture 3DGS with?
Any sufficiently thorough set of photos or video can train a splat scene, but ad-hoc capture makes quality a lottery. Purpose-built hardware makes it repeatable. Rent the PortalCam, a dedicated Gaussian splatting camera, when the output is purely visual and capture simplicity matters most: walkthrough capture, one workflow, splat scene out. Rent the Lixel K2 when the project needs geometry and visuals: its LiDAR and cameras record together, so one walk produces a point cloud with 1 cm relative accuracy for the technical team and a 3DGS scene for everyone else. In both cases processing runs through LCC CyberColor, XGRIDS's GPU-based 3DGS pipeline.
Gaussian splatting for real estate and virtual tours in Singapore
The obvious first market is property: walkable, photoreal scenes of showflats, resale listings, and event venues that a buyer can explore from a phone. Compared with hosted tour platforms, the 3DGS rental route has different economics: capture per property with rented hardware, own the output, no per-listing platform subscription, and the visual result is a genuine 3D scene rather than linked panoramas. The K2 vs Matterport comparison sets the two approaches side by side. Beyond real estate, the same workflow serves venue marketing, media and virtual production backdrops, heritage records, and the presentation layer of digital twins: the scene a stakeholder actually enjoys navigating, backed by point cloud data when captured on the K2. What a rental covers for this workflow is on the 3DGS spatial modelling page.
How do you capture good splats?
- Light the space as you want it remembered. 3DGS faithfully reproduces whatever lighting you give it; in Singapore, tropical light changes fast, so capture each space in one session for consistent appearance.
- Cover viewpoints generously. The reconstruction can only know what the capture saw; move steadily and vary height and angle around objects of interest.
- Mind dynamic scenes. Moving people and vehicles become ghostly artefacts; schedule capture for quiet windows: showflat and venue access windows are usually short, and walkthrough capture fits them.
- Plan loop closures on the K2. The 3DGS output inherits the quality of the underlying SLAM trajectory, so the habits from handheld LiDAR mapping apply unchanged.
What does 3DGS capture cost?
Volumet's weekly dry-rental rates in Singapore, GST-exclusive: PortalCam S$1,200, Lixel K2 S$2,300. First rentals add a one-time onboarding session, S$250 if you self-collect, S$400 delivered to site, covering device handling and the capture-to-CyberColor workflow. A week comfortably covers multiple properties or venues at walkthrough pace, so per-scene cost falls quickly with volume: four properties on a PortalCam week is S$300 each before processing time. Availability for your dates is on the rental page.
Limitations, honestly
3DGS is not a measurement format: it is built to be seen, not dimensioned, and modelling stays in point cloud territory. Quality depends on capture coverage, lighting, and processing; mirrors and large glass surfaces remain hard for any image-based method; and very large scenes need capture planning and segmentation. None of these are reasons to avoid the format: they are reasons to pick the right device. Teams that need both deliverables should capture with the K2 rather than chase splats alone; teams that only need the visual should take the PortalCam and the simpler workflow. Tell us the deliverable and your dates and we will recommend the right kit and confirm a quote.