Why we build on 3D Gaussian Splatting, not point clouds
A point cloud tells you where the surfaces of a building are. A splat tells you what the building actually looks like. On a compliance record, that difference is the whole point.
Most of the scanning industry still hands clients a point cloud: millions of individual coordinate points, each with a colour value, floating in space. It is a genuinely useful format for measurement, and it has been the standard for a long time. It is also, frankly, not something anyone would choose to look at. Gaps show through walls. Reflective and glass surfaces come back as noise. Nobody mistakes a point cloud for a photograph.
What a splat actually is
3D Gaussian Splatting represents a scene as millions of small, soft, coloured shapes, each one a tiny blurred blob with its own position, size, orientation and colour, rather than a hard point or a triangle in a mesh. Rendered together at scale, those shapes reconstruct light and surface detail with a fidelity that point clouds and meshes cannot match, and they render in real time in a standard web browser, with no dedicated software and no headset required.
The practical result on a live twin is that reflective plant, cabling, signage text and fine surface texture, the things that tend to matter most on a compliance walk or a claims survey, actually survive the capture process instead of dissolving into noise.
Why fidelity is not just cosmetic
This matters for more than appearance. Detection models look for visual signal, a crack, a stain, a missing guard rail, and a model can only find what the underlying capture preserves. A twin built to look convincing from a distance but blurred up close will quietly fail at the one job that pays for it: catching the thing a human would have caught on site.
OnXR runs a four level quality system across every capture, Preview through to Ultra, so a team can choose fast turnaround for a routine walk and reserve the highest fidelity setting for the areas that will end up as evidence, a defect, an incident, a claim.
Worth being honest about: higher fidelity costs more processing time and more storage. We do not default every scan to the top setting because most of a building does not need it. Judgement about where fidelity matters is part of what a proper capture plan gives you, not something the software decides alone.
Capturing that data starts on site with a handheld scanner, and turning a raw scan into a usable twin is its own process. The next post walks through it end to end.
See the fidelity difference for yourself
We can show you a splat twin of a real site next to a traditional point cloud of the same space. The difference is obvious within thirty seconds.
See the product