3DGSUE5VFXWorkflow 2026-07-24

Exporting Assets from 3DGS Scans
with UE5 × XGRIDs SDK

I'm Kou Nakamura, founder of Locahun 3D. If you want to use 3DGS scans as real production assets, nothing beats bringing them into a game engine. In this article I walk through the UE5 workflow I actually use with my scan of the Shibuya scramble crossing: installing the XGRIDs SDK (LCC Plugin for UE), exporting assets with a virtual camera, and changing the lighting with AI. I've also included comparisons of the major 3DGS tools and the other UE5 plugins that can handle splats.

The Shibuya scramble crossing 3DGS scan loaded into a UE5 viewport, ground-level angle, daytime
The Shibuya scramble crossing 3DGS loaded into UE5. Importing the LCC2-format scan with LCC Plugin for UE lets you walk around the viewport freely while hunting for angles.
3DGS scan (Shibuya) Import via LCC Plugin for UE Lock the framing with a virtual camera Export stills / video Relight with AI

01 Why bring 3DGS into UE5 at all

At Locahun 3D I produce and sell 3DGS scan data, mostly of urban locations. A splat viewer is fine for scouting, but if you want to use a scan seriously as production material, importing it into UE5 unlocks a whole extra tier of freedom. Concretely, you get to:

A CG sports car composited onto the street of the Shibuya 3DGS scan
Compositing a CG asset into the scan. A CG car model placed on the street of the Shibuya 3DGS scan. Because the base is a real-world scan, perspective, scale, and lighting stay consistent, which makes VFX blocking fast.
Image by senior VFX artist hectorVFX, using Locahun 3D's Shibuya data.

The downstream step — finishing the exported frames into aerial plates — will get its own article, so this one stays focused on the UE5 integration and asset export. The plugin I use is XGRIDs' LCC Plugin for UE (free).

02 What is XGRIDs' "LCC Plugin for UE" (and where to get it)

LCC Plugin for UE is the official UE5 3DGS plugin distributed by XGRIDs. It handles loading, rendering, editing, and relighting of 3DGS scenes in a single plugin. I use it for two main reasons: it scales to city-sized scans (streaming LoD up to billions of points) and it supports LCC2, a heavily compressed format. For a large scene like Shibuya, the files are dramatically lighter than passing PLY around, and loading stays stable.

Engine support
Unreal Engine 5.1–5.8 (Windows / Linux, DirectX 11 & 12 / Vulkan)
Formats
LCC2 / LCC / PLY / SOG / SPZ. Standard 3DGS PLY files from Postshot, Luma, Polycam, and similar tools load directly
Scale
No point-count limit. Streaming LoD keeps frame rates stable even in huge scenes
Pricing
Free + Pro tiers. Pro is currently free for a limited time (adds proxy-mesh relighting, self-shadowing, ACES/OCIO color pipeline, and more)
Usage
Blueprint-based, no code required. A C++ API is available for runtime control
Download / Official XGRIDs

LCC Plugin for UE (plugin + sample data)

The plugin ships as a per-UE-version zip from XGRIDs' developer portal. Sample data is on the same page.

https://developer.xgrids.com/#/download?page=LCC_UNREAL_SDK_UE54

Open the download page →

03 Installation

Installation follows the standard UE5 plugin flow. I run LCC4Unreal-v3.0.0-win-UE5_5 in a UE5.5 project.

  1. From the download page above, grab the zip that matches your exact UE version. A version mismatch is the classic cause of load errors, so this is the one step to get exactly right.
  2. Extract the zip and place it in the Plugins folder at your project root (create the folder if it doesn't exist). If you want it available to every project, you can put it in the engine-side Engine/Plugins instead.
  3. Launch the project, open Edit → Plugins, search for "LCC", and enable it. The editor asks for a restart — go ahead and restart.
  4. After the restart, use the plugin's file dialog to load a .lcc / .lcc2 / .ply file, and the 3DGS scene is placed into your level.
About exact menu locationsThe plugin's UI and menu placement can change between versions. XGRIDs' official tutorials and developer docs are the primary source for the current steps. If you hit a wall during setup, check there first.

04 Load the 3DGS and export assets with a virtual camera

Once the scan is loaded, everything else is standard UE5. Move the camera in the perspective view and build the framing you want. For the Shibuya scramble crossing, a straight-down view from above the station can't be shot in reality — it's a drone no-fly zone — but inside the scan there are no restrictions. The core of this workflow is using a real ground-level scan as the foundation, then freeing only the camera from physics.

A straight-down aerial angle of the Shibuya scramble crossing created with a UE5 virtual camera, daytime
A straight-down angle built with the virtual camera. Moving the camera directly above the station produces a framing no drone is allowed to shoot. The crossing paint and crosswalk detail all come from the real scan.

With the framing locked, export it as an asset. For stills, running a High Resolution Screenshot (HighResShot) at a specified resolution gives you a frame ready for compositing and retouching. For video, building a camera animation in Sequencer lets you render the shot with full camera movement.

Key pointThe foundation is always the actually-scanned 3DGS data. You're not asking AI to generate a background from scratch — you're adding only a "physically impossible camera move" on top of a photo-based point cloud. Composition, building placement, and scale accuracy are all locked at this stage.

05 Changing lighting and season with AI

Hand the stills exported from UE5 to an image-generation AI, and swap the lighting and season with nothing more than a short prompt like "make it night" or "make it snow." Since composition, camera angle, and building placement are already fixed on the UE5 side (i.e. by the real scan), the AI only has to handle light and atmosphere. The shorter the prompt, the better the composition holds — this division of labor produces the fewest artifacts.

Ground angleThe source is on the left. The framing stays fixed; only the time of day and season are swapped.
Source Source still. The Shibuya scramble crossing exported from the 3DGS scene in UE5 (ground angle, daytime)
Still exported from UE5 (daytime). The framing of the raw scan.
AI “make it night” A night view of the Shibuya scramble crossing generated with the prompt 'make it night'
Night. The AI fills in window lights, glowing signage, and road reflections. The composition matches the source.
AI “make it a snowy night” A snowy night view of the Shibuya scramble crossing generated with the prompt 'make it a snowy night'
Snowy night. Only the road surface, street trees, and sky change.
Straight-down angleThe same steps work on any framing. The source here is the top-down shot built in 04.
Source Source still. The straight-down angle of the Shibuya scramble crossing built with a virtual camera in UE5 (daytime)
Straight-down, built with a virtual camera (daytime). An angle you cannot fly a drone for.
AI “make it dusk” The straight-down Shibuya angle changed to dusk lighting with AI
Magic hour. It reads as an aerial-style shot.
AI “make it night” The straight-down Shibuya angle changed to night with AI
Night. Locking one angle lets you line up time-of-day variations side by side in minutes.

06 Current limits: signage, text, and man-made detail break down

To be honest about the constraints: signage text and fine man-made structures — trains, road signs — break down under current AI completion. It produces "ghosts" (structures that shouldn't exist) and artifacts like train-shaped geometry generated where no train belongs.

Annotated image marking AI completion artifacts: ghost structures, an implausible generated train, and broken signage text, circled in red, yellow, and blue
Examples of breakdown. Yellow circle = a structure that shouldn't exist (GHOST). Blue line = train-like geometry generated where no train belongs (TRAIN). Red circles = signage and road-sign text falling apart.

I consider this exactly where professional work lives. By planning reshoots and cleanup passes for the fragile areas in advance, you can bring the result up to deliverable quality. As for the AI completion itself, I expect the accuracy to keep improving.

When to use itThis technique is not universal. It suits framings without signage, text, or fine man-made objects, and shots where minor artifacts are acceptable. For shots where sign text must read accurately, it's currently the wrong tool.

07 Handling 3DGS in existing DCC software

Game engines aren't the only option — the DCC software already in your VFX pipeline can now handle 3DGS directly. To decide how much to do in a DCC versus in UE5, here's where the major packages stand.

Software3DGS supportStrengthsWeaknesses
Houdini 22NativeTreats splats like regular geometry (point clouds) — deform and animate them with standard nodes like Bone Deform. Renders directly in Karma, relights in Copernicus. Unmatched freedom for procedural splat workLicense cost and learning curve are steep. Overkill if you just want to view a scan
BlenderVia add-onsFree add-ons like KIRI Engine's "3DGS Render" cover import, editing, and rendering at zero costAdd-on dependent, so features and quality vary; heavy with large splat counts
3ds MaxVia V-Ray 7V-Ray 7 renders Gaussian Splats directly — slots naturally into existing V-Ray pipelines (archviz and beyond)Requires a V-Ray license; splat editing is limited
Cinema 4DVia renderersRenders 3DGS through third-party renderers such as Octane. Pairs well with camera work and motion graphicsNot a native C4D feature — depends on your renderer and its version

My own split: UE5 for walking through city-scale scenes and hunting angles, the DCC side for per-shot look development and rendering. For aerial-plate work I use a C4D → Octane flow, walked through in detail in Generating an "aerial shot" from 3DGS. For procedural splat work in Houdini, see the Houdini 22 × ComfyUI hybrid workflow article, where hectorVFX demonstrates it on the same Shibuya data.

08 UE5 3DGS plugins that can relight

There are plenty of UE5 plugins that can merely display 3DGS, including free and open-source options (XScene-UEPlugin, MLSLabs GS Renderer, NanoGS). But once you're exporting assets like in this article, the practical dividing line is whether you can change the lighting of an imported splat scene. So here I'm covering only the plugins that support relighting.

PluginPriceRelightingStrengthsCaveats
LCC Plugin for UE
(XGRIDs — used in this article)
Free + ProBasic relighting in the free tier; proxy-mesh relighting + self-shadowing in ProStrongest for whole-city scenes, with billion-point streaming LoD and LCC2 compression. Also reads PLY/SOG/SPZPro is free for a limited time; check future pricing
Volinga PluginFree + ProFree tier: additive lighting that interacts with scene lights. Pro: proxy-mesh relighting reproducing geometry, shadows, and materialsBuilt for professional virtual production — multi-GPU with nDisplay, ACES/HDR, disguise integration. Reads PLY and its own NVOL formatPro features are paid; the design is optimized for LED-wall VP work
Luma AI UE PluginFreeRelighting via analytical and sun lightsTakes Luma captures straight into UE, with free crop, cull, and merge toolsRelighting covers direct light only (no emissive); supports UE 5.1–5.3

One caveat shared by all three: splats are rendered by a custom pipeline that sits beside Unreal's core renderer, so they don't participate in Lumen global illumination. Relighting in every case works through proxy meshes — simplified geometry fitted to the splats — to reproduce shadows and light response. Within that field, I've settled on the XGRIDs SDK because it combines relighting support with the LCC2 format design that lets an entire city block live in a single scene.

09 Summary

The whole flow in one line: import the 3DGS scan into UE5 with LCC Plugin for UE, lock the framing with a virtual camera, export stills and video, and let AI swap only the light and season. The essential rule is to never let AI generate everything — always route through the real scan CG. Composition and accuracy are guaranteed by the scan data and the virtual camera; AI only does the finishing. That division of labor is what makes "impossible shots" usable at production quality with minimal artifacts.

If you're interested in UE5 integration or using 3DGS as VFX material, feel free to get in touch.

DATA / 3DGS

The Shibuya scramble crossing 3DGS data used in this article is
available on Locahun 3D Online

Provided in PLY / OBJ format with a 3DGS walkthrough. Standard license from ¥200,000 (commercial and non-commercial use). If you'd like to try the same UE5 × AI workflow on the same data, take a look.

View the property data →
Credits

Credits

Kou Nakamura
Kou Nakamura中村 航
Founder, Locahun 3D / Research & Writing

Founder of Locahun 3D (LOCAHUN 3D). Researches and validates production workflows that combine 3DGS scanning with game engines and AI.

Follow @Kou45388803 →
hectorVFX
hectorVFX
Senior VFX Artist / Composite figure

Senior VFX artist based in Budapest, Hungary. The CG-car composite example in this article is hectorVFX's work, created with Locahun 3D's Shibuya data. His Houdini 22 × ComfyUI workflow is covered in detail in its own article.