06

CAPTURE / 06 — 3D GAUSSIAN SPLATTING

3D Gaussian Splatting

Photoreal, real-time radiance fields of sets, environments, and objects. A splat scene renders like a photograph you can move through — and every set and location we scan is captured to be processed into 3DGS.

[ 3DGS-READY CAPTURE / REAL-TIME SPLATS ]

Photoreal 3D reconstruction of a high-desert canyon rendered from SCANable aerial capture

SRC / RADIANCE FIELD / PHOTOREAL, REAL-TIME

WHAT IT IS

A photograph you can walk through.

3D Gaussian Splatting reconstructs a real place or object as a dense field of soft, overlapping gaussian “splats” — each one carrying position, color, and view-dependent shading. The result is a radiance field that renders in real time at photographic fidelity: reflections, soft edges, and subtle lighting that a traditional mesh-and-texture asset can only approximate, at a fraction of the runtime cost.

For film and episodic work that means a scouted location, a struck set, or a hero prop can live on inside a real-time engine exactly as it looked on the day — explorable for previs, virtual art department, LED-volume backgrounds, and interactive.

The same canyon split down the middle — photoreal radiance field on one side, the underlying reconstruction on the other FIG.06 / SAME CAPTURE — FIELD LEFT, GEOMETRY RIGHT

CAPTURED 3DGS-READY BY DEFAULT

Every set scan is a splat waiting to happen.

This is the part that sets us apart: 3D Gaussian Splatting isn’t a separate shoot. Every set and location our crews scan is captured photogrammetrically — dense, overlapping, color-calibrated coverage from enough angles that the same data can be reconstructed into a splat scene later, without sending anyone back. When you book set or location scanning, you’re already banking 3DGS-ready capture.

So the decision to deliver a Gaussian splat doesn’t have to be made on set, under time pressure. Scan the environment the way we always do, and the option to turn it into a real-time radiance field stays open for the length of the production.

[ PHOTOGRAMMETRIC COVERAGE / COLOR-CALIBRATED / RECONSTRUCT ANYTIME ]

ONE SET SCAN, TWO DELIVERIES

A scanned interior lobby rendered with full photographic texture
01 — Photoreal, as captured
The same lobby shown as untextured reconstruction geometry
02 — Underlying geometry

HOW SCANABLE DOES IT

Capture once, reconstruct to spec.

Our crews shoot the dense, overlapping photogrammetry a clean splat reconstruction needs, run it through our color-calibrated RAW pipeline, and train the radiance field to the fidelity and file budget your engine expects. We tune density, clean floaters, and lock scale so what you load is a usable scene — not a raw capture you have to salvage.

Because our crews mobilize from six cities, 3DGS capture happens on your production — a backlot, a practical location, or a soundstage before strike — anywhere in the country.

A full stadium bowl reconstructed photorealistically at true scale, seating and structure holding detail across the whole environment

SRC / LARGE-SCALE ENVIRONMENT / RECONSTRUCTED TO SCALE

WHAT YOU RECEIVE

Splats your engine can open.

  • Gaussian splat scenesTrained radiance fields (.ply / .splat) of environments and objects.
  • Engine-ready deliveryTuned for real-time playback in Unreal, web, and interactive viewers.
  • Scale-locked & cleanedTrue-scale, floater-free scenes registered to your set data.
  • Source coverage retainedThe photogrammetry stays on file for re-training or mesh delivery.

START HERE

Turn a location into a real-time scene.

Tell us the environment and the engine it has to run in. We’ll scope the splat capture — or pull it from set data we’re already shooting.

Request a Quote

FAQ

Common questions

Does genlock actually matter for Gaussian splatting?

Yes. If the cameras are not genlocked, each one is sampling a slightly different moment, and any reconstruction of a moving subject inherits that error. Every array we build is genlocked for that reason.

What is the difference between 3D scanning and 4D capture?

A 3D scan is a single instant — one frame of geometry and texture. 4D capture is performance over time: the same volumetric capture repeated across a sequence, so motion and deformation are captured, not simulated after the fact.

What is C.A.L.V.I.N.N.™?

A separate modular array built in house for volumetric and machine-learning capture, including 4D Gaussian splatting. It is a distinct system from the Mobile CyberSCAN™ Studios, which handle static scanning and OLAT.

What do you deliver?

That depends on the pipeline you are handing it to — we scope deliverables with your VFX vendor before the shoot so the data lands in a form your team can use on day one, not after a conversion pass.