Imagine capturing a place or object and returning to it whenever you like. Not as a memory, but as an immersive scene you can move through. You can study surfaces up close and inspect every detail. Light falls where it should. Textures and depth feel real. This is the level of fidelity the industry has been chasing for decades.
Let's start with something familiar. You have seen images like this many times. But what if an image could become something more? Gaussian splatting offers a new way to represent objects and scenes, turning flat pictures into navigable 3D spaces with depth, volume and presence. So how does it work?
The first step is capture. You walk around the scene and film it from different positions, or send a drone overhead. The camera keeps moving, collecting hundreds or thousands of views. Each one shows the same place from a slightly different angle.
Computer vision algorithms analyse the images and estimate how the different camera views relate to one another in 3D space. From this, the system reconstructs an initial sparse structure of the scene and begins optimising millions of Gaussian splats to match the original photos. Unlike neural rendering systems that require a heavy model at runtime, Gaussian splatting stores the optimised scene directly as splats, enabling real-time rendering on modern GPUs.
The system starts with rough points in space and tests them against the original images again and again. Some splats are removed, while others are split, refined, or expanded as the optimisation progresses. Over tens of thousands of iterations, the model keeps checking what matches, what does not, and how the structure should be adjusted.
A splat is a small 3D shape with soft edges. It is strongest at the centre and fades out as it spreads into the surrounding space. That fade follows a mathematical pattern known as a Gaussian, named after the 19th-century mathematician Carl Friedrich Gauss.
Each splat carries a small bundle of information: its position, size, direction, colour and transparency. On its own, it is almost nothing. In large numbers, they begin to form continuous surfaces and volumetric structures. Millions of splats combine into recognisable forms.
People working with splats notice the same pattern: the technique opens doors that older methods kept shut.
Gaussian splatting unlocks what we always struggled with: cables, foliage, and transparent elements such as haze or glass. It is already impacting the industry, and the demand is rising.”
Researchers introduce 3D Gaussian splatting
Polycam launches Gaussian splatting tool
Superman brings Gaussian splatting into a feature film pipeline
Search interest in Gaussian splatting reaches a new high
Researchers introduce 3D Gaussian splatting
Polycam launches Gaussian splatting tool
Superman brings Gaussian splatting into a feature film pipeline
Search interest in Gaussian splatting reaches a new high
Search interest over a specific time period, displayed on a relative scale from 0 to 100, where 100 signifies the peak interest for the time period of the chart. A value of 50 indicates half the popularity of the peak and 0 suggests insufficient data.
© Méridien 2026 — CC BY-NC 4.0. According to Google Trends Data:
Web Search, worldwide, period 5 years, 15.04.2026
Search interest over a specific time period, displayed on a relative scale from 0 to 100, where 100 signifies the peak interest for the time period of the chart. A value of 50 indicates half the popularity of the peak and 0 suggests insufficient data.
Gaussian splatting is on the rise. Architects use it for quick and convincing previews: a bare concrete shell can become an immersive walk-through in a matter of hours. Retailers are experimenting too, turning products into interactive 3D objects that can be explored from any angle or dropped into a living room through AR.
For museums, splatting allows for capturing fragile rooms and artefacts without touching or moving anything. Entire galleries or historic houses can be turned into explorable digital spaces. At the same time, delicate objects — textiles, instruments, jewellery — can be preserved in full detail. It works both as a conservation tool and a way to share collections far beyond the museum's walls.
And the media is beginning to pick it up. Immersive 3D scenes help readers grasp complex stories more intuitively. The New Yorker has already published its first splatted scene, and The New York Times is experimenting too.
Capturing reality
3D Gaussian splatting
Knowledge and context
Verifiable knowledge
Spatial information
Future technologies
Machine learning
AI agents
Active engagement improves retention. A live dialogue with an LLM turns learning into exploration. Users can ask questions as they move through a place or model and receive answers linked to what they are seeing. The result is a guide that feels personal, but can scale to millions.
Building a 3D scene is only the starting point. To make it useful, the scene needs context: what things are, where they are, how they relate, and why they matter. This is where large language models and AI agents come in. The scene becomes a spatial interface linked to structured, verifiable knowledge. A construction site, a building, or a collection becomes something you can navigate and understand.
At Méridien, we shape that knowledge layer. Our technology partners at Datum build the system behind it, combining 3D Gaussian splatting, spatial information, machine learning, verifiable knowledge, and AI agents into a single platform. Each element in the model can be linked to data, turning the capture into a dynamic source of knowledge.
In tourism and museums, this means immersive guides. For corporate training, it provides structured 3D environments for learning. Construction gets spatially aware monitoring and control. From just capturing what something looks like, we move to truly understanding what it is and how it works.
By enhancing 3D Gaussian splatting with AI, we turn static models into interactive environments where people can explore, ask questions, and get useful answers in context.
2 GB
With Gaussian splatting, a scene built from 2 GB of raw data loads in just a few seconds.
A construction site, a desalination plant, or a solar power station becomes easier to inspect, explain, and learn from.
Encyclopaedias, YouTube channels, and lectures deliver information passively, without answering follow-up questions or adapting to the learner's level. Datum is a pioneering immersive technology company specialising in AI-powered digital tours and spatial storytelling through advanced visualisation techniques. Its flagship platform, Datum Teller, transforms how institutions and industries communicate spaces, stories, and experiences by combining technologies such as 3D Gaussian splatting, data-driven narratives, and interactive environments.
By enhancing 3D Gaussian splatting with AI, we turn static models into interactive environments where people can explore, ask questions, and get useful answers in context.
A construction site, a desalination plant, or a solar power station becomes easier to inspect, explain, and learn from.
Capture cities, infrastructure, and public services as up-to-date 3D records. These models can support planning, maintenance, emergency response, and transparent decision-making.
Document cultural sites, events, and performances as immersive 3D records. Audiences can revisit exhibitions, explore venues, and experience works remotely, extending access beyond physical limits.
Turn lessons into interactive 3D environments. Students can explore spaces, examine objects, and ask questions in context, rather than passively reading or watching.
Capture sites regularly to monitor progress over time. Compare current conditions with plans, document changes and share updates with teams and stakeholders.
Capture roads, utilities, and public assets in an immersive 3D. Teams can inspect conditions, track changes, and plan maintenance without repeated site visits.
Record complex facilities such as plants and refineries as detailed 3D environments. Workers can inspect equipment, review layouts, and plan operations remotely.
Published
Reading list
Osama Jamil, AnnMarie Brennan. Immersive heritage through Gaussian Splatting: a new visual aesthetic for reality capture. Frontiers in Computer Science, 2025.
Michael Rubloff. Gaussian Splatting at the New Yorker. Radiance Fields, 2025.
Andy Gstoll. Explore The Magical 3D Gaussian Splats Museum. Mixed Reality, 2025.
AJ Chavar et al. Pushing the Limits of Gaussian Splatting for Spatial Storytelling. The New York Times Research & Development, 2024.
Char Stiles. Splat Sketches. MIT Media Lab, 2024.
Stanford Lee et al. 3D Gaussian Splatting: Performant 3D Scene Reconstruction at Scale. AWS Spatial Computing Blog, 2024.
