C-DRONE GUIDE · 4 SEPTEMBER 2026

Drone Gaussian splatting: the new 3D rendering technique, how it differs from photogrammetry, price

Since 2023, a technique born in computer-graphics research has been changing how a drone survey gets turned into 3D: Gaussian splatting. Instead of a fixed mesh or point cloud, the scene is rebuilt as a cloud of millions of small, semi-transparent shapes optimised to match the captured photos from every angle — rendered photorealistically, in real time, viewable in a browser with no specialist software. That is enough to appeal to a property developer who wants to let people tour a site remotely, or a local authority wanting to showcase a heritage site. But the technique has a trade-off few providers explain clearly: a splatting render is not a metric survey in the sense a surveyor or engineering firm means it. Here is what the research says, what the technique genuinely adds for a professional, and where it stops.

Published on 4 September 2026, reviewed on 11 September 2026 — regulations in force as of September 2026.

What Gaussian splatting changes compared to a classical 3D model

Classical photogrammetry, used for years for drone surveys, rebuilds a scene in two steps: it first computes a point cloud from the overlap between photos (structure-from-motion), then generates a textured mesh — a surface made of triangles, with the original photo mapped onto it. Every point in the cloud, every vertex of the mesh, corresponds to a real 3D coordinate, tied to GNSS ground control points when the survey needs to be georeferenced.

Gaussian splatting (3DGS) starts from a different principle, first described in 2023 by a research team spanning Inria, Université Côte d'Azur and the Max Planck Institute for Informatics — Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler and George Drettakis, in a paper that has become the field's reference, published in ACM Transactions on Graphics for SIGGRAPH 2023 (see the study on Google Scholar). The scene is represented by millions of small, blurry three-dimensional shapes — "Gaussians", each defined by a position, orientation, size and colour — whose parameters are optimised through learning so that, overlaid and viewed from any angle, they faithfully reproduce the captured photos. The result renders in real time, above 30 frames per second, with a far more photorealistic look than a classical mesh: reflections, depth blur, dense foliage or water surfaces — subjects that traditionally give poor results in photogrammetry — come out convincingly.

What the research says: impressive on screen, imprecise on the ground

The point many commercial demos skip over: a Gaussian is not anchored to a precise point on a real surface, unlike a photogrammetric point cloud. The model optimises a faithful visual render from the training photos' viewpoints — not an exact geometry. A study published in 2025 in the journal Sensors by Muhammed Enes Atik, of the geomatics department at Istanbul Technical University, compared point clouds extracted by Splatfacto (the Gaussian-splatting implementation in the open-source Nerfstudio software) against a classical photogrammetric reference, using drone imagery over two study areas: the measured horizontal deviation ranges from 0.2 to 1.35 m depending on the area (see the study on Google Scholar). In return, processing is markedly faster: convergence in about fifteen minutes for Splatfacto, against 45 to 60 minutes for the neural radiance field (NeRF) rendering methods compared in the same study.

In practice, for a professional, that draws a clear line: Gaussian splatting is reliable for showing a site — an immersive tour, a communication image, a presentation model — but not yet for measuring it with the rigour a materials cubature, a settlement or crack-monitoring survey, or a plan usable by an engineering firm requires. For those uses, our guide to drone photogrammetry and 3D modelling details the classical method, which remains the georeferenced reference.

Professional use cases: where splatting adds real value

For communication and showcasing, splatting has an edge classical photogrammetry cannot match: an already photorealistic render, with no texture-retouching step, viewable in a browser on a computer or phone, with no plugin or specialist software. Three concrete professional uses draw on that.

Off-plan property marketing and sales. A developer can give a remote, immersive tour of a site under construction, or of its setting (neighbourhood, views from upper floors), to a buyer who cannot visit in person — a complement to classical CGI renders, with the advantage of a real site rather than a simulated image. Our guide to drone use for marketing a new-build property scheme covers the related uses (photo, video, periodic site-progress tracking).

Showcasing a heritage or tourism site. A tourist office, a local authority managing a monument, or a wine estate can offer an immersive virtual tour of a site that is hard to access (a roof, a listed interior, private land) — an outreach use rather than a technical diagnosis, to be distinguished from a roof survey, which remains photogrammetry's domain: see our guide to drone roof inspection of a listed monument.

Site-progress communication for stakeholders. A project owner, architect or construction firm can share an immersive progress render with elected officials, residents or a steering committee, without waiting for a full BIM digital twin to be processed — useful alongside, not instead of, the drone-based digital twin of a building or site, which remains the reference tool for technically exploiting the model.

Method, tools and 2026 prices

The drone capture itself is identical to a classical photogrammetry mission: high overlap between photos (often higher than a simple orthophoto requires), crossed nadir and oblique passes to cover façades. It is the processing that differs: instead of — or alongside — classical photogrammetry software, the photos are optimised by a tool dedicated to Gaussian splatting (from open-source solutions such as Nerfstudio to features built into some commercial drone-processing software). The resulting file is then viewed through a web viewer or a dedicated app, without requiring the computing power of a classical 3D workstation on the viewer's side.

Ranges observed in France in 2026, excl. VAT:

The right approach for a professional is to state the intended use in the quote request itself: if only a visual render is needed, splatting alone is enough and costs less than a full survey; if the render must also support a measurement, a cubature or a technical file, the mission should include a classical georeferenced photogrammetry component — a splatting render can still be derived afterwards from the same photo set for communication purposes.

The information on this page reflects the state of the technique and the rules in force in September 2026; the flight remains subject to the same rules as any drone photography mission (category, authorisations depending on location). For a developer, a local authority, a tourist office or a project owner, request a quote stating the site, the intended use (communication or measurement) and the desired timeframe.

Request a free quote

Put it into practice

• Drone surveying & photogrammetry: rates and cities covered from €800
• Surveying & photogrammetry in Valence Auvergne-Rhône-Alpes
• Surveying & photogrammetry in Lorient Bretagne
• Surveying & photogrammetry in Bayonne Nouvelle-Aquitaine

Also worth reading

• Olive orchards and drones: water stress and yield forecasting, price
• Prune d'Agen and drones: water stress and ripeness, price
• Asparagus and drones: spotting weeds on the ridge before emergence, price