Gaussian Splats don't replace point clouds … and that’s the point!
There’s been a lot of noise around Gaussian Splats lately. Some people are excited; others are dismissing them entirely.
The main criticism I keep hearing is: “They’re not accurate enough for reconstruction or 3D modelling.” And that’s true — if you expect them to behave like a laser scan or a CAD model.
But that’s not what they are. Gaussian Splats don’t produce clean, editable geometry. You can’t snap to sharp edges. You can’t extract perfect linework. You can’t remodel from them the way you would from a structured point cloud.
If your goal is millimetre-level measurement, layout, clash detection, or BIM modelling, you still need proper geometry. That hasn’t changed.
A Different Problem to Solve
Where I think the conversation goes wrong is that we’re judging splats as if they’re supposed to replace survey-grade data. They’re not. They solve a different problem.
Instead of focusing purely on precision, splats focus on presence. They capture how a space actually feels — the light, the texture, the depth, and the complexity of real environments. They preserve context in a way that’s immediately understandable, even to someone who has never been on site. And that matters more than we sometimes admit.
In our workflows, we don’t choose between point clouds and splats — we layer them. The point cloud gives us dimensional honesty. It’s the backbone. It’s what we use to check clearances, verify alignment, and ensure the design physically works.
The splat model gives us immersion. It allows teams to walk through the space virtually. It helps stakeholders understand scale. It keeps remote decision-makers aligned on what the site is actually like. One supports engineering accuracy; the other supports human understanding — and both are necessary.
From Data to Confidence
When we combine a concept model from SketchUp with captured site context and visualise it through tools like V-Ray or D5 Render, the discussion changes. It stops being just about “Is it technically correct?” and becomes “Can we see what this will really be like?”
That shift is powerful because, in construction, the real deliverable isn’t a dataset — it’s confidence. Clients don’t just want measurements. They want clarity. They want to understand what changes, what stays the same, how it affects the real world, and whether the scale feels right before they commit time and money.
If the goal is designing, you still need clean geometry. If the goal is believing, splats dramatically reduce the gap between information and understanding. So maybe the better question isn’t “Are Gaussian Splats accurate enough?” it’s “What decision are we trying to help people make?”
Used correctly, splats aren’t hype. They’re a new layer of communication in an industry that has traditionally struggled to make complex information easy to see.
And in my view, that’s where their real value lies.
– Chris Keightley | BuildingPoint SA (Pty) Ltd