VFX
Capture is only half the story. Once a performance has been recorded on our VCR rig, the real craft happens in post — turning that raw volumetric data into a digital asset that can live in any scene, under any light. This is where our VFX partner takes over: relighting Gaussian splats to the level of a fully integrated, production-ready shot.
A Growing Professional Toolset
Gaussian splatting has moved fast from research curiosity to an established part of the professional VFX pipeline, and the range of pro-grade software supporting it now is genuinely broad. Depending on the shot, the platform and what the project needs, we work across:
- Houdini — using GSOPS, a dedicated Gaussian splat toolset, to combine 4DGS with traditional 2D/3D VFX work
- Unity — combined with a render engine like the bespoke Deckard Render pipeline we used for the boxer scene, for photoreal rendering with true lens characteristics: depth of field, bokeh, focal behaviour
- Unreal Engine — for real-time visualisation, virtual production and in-engine work with splat data
- Nuke — for final compositing, grading and integrating splat renders into finished shots
It's not one fixed pipeline — it's a toolkit, and we choose whichever combination of these gets the best result for the shot in front of us.
Can I Relight Splats In Post
A standard Gaussian splat carries the lighting it was captured under baked into the data — great for playback, but limiting the moment you need that performance to sit inside a different environment. The VCR rig workflow breaks that link. The capture is processed so it can be re-lit after the fact, meaning a performer captured once on the VCR rig can be placed convincingly into daylight, firelight, a moody interior, a CG environment — whatever the shot calls for — without stepping back in front of a camera.
The result is a genuinely relightable 4D Gaussian splat: full 360°, full motion, and no longer tied to a single lighting setup.
Why It Matters For Production
- Photoreal quality from every angle — including the fine detail traditional CG struggles with: hair, fabric, skin, complex materials
- Performance captured directly from the actor, not hand-animated
- One capture session, reused across multiple shots, sequences and environments
- Segmentation tools allow individual elements — clothing, props — to be isolated and varied, giving multiple looks or character variants from a single capture
- Far less reliance on manual modelling, texturing and simulation
- Faster turnaround, lower cost per shot than a from-scratch digital double
Where It's Used
This pipeline is built specifically around digital humans and crowd work — digital doubles, crowd replication, and performance capture for virtual production — as well as environment, prop and location capture for VFX plates. It's already been deployed on major feature productions, handling scenarios that would traditionally require full CG builds.
Built On Machine Learning, Not Generative AI
Gaussian Splatting is a machine learning technique, not generative AI — every result is derived directly from the footage captured in the rig, with no synthetic or generated content introduced. Processing runs through machine-learning-powered tools purely to refine quality and optimise the data, within a closed, secure, fully audited production pipeline suitable for NDA-sensitive shoots.
PRISM AI LTD — LONDON