For those who don’t know the tool: The Pixel Farm makes PFTrack, a node based matchmove and reconstruction tool that exports to DCCs like Autodesk Maya, Foundry Nuke, SideFX Houdini, Blender and Unreal Engine, with improved dense point clouds via OpenUSD.

Hero Cloud is a new reconstruction node

PFTrack Build 26.05.19 adds Hero Cloud, a node that generates a measured dense 3D point cloud from a single tracked plate. It sits in the Geometry category and runs downstream of a solved camera, turning an ordinary camera solve into spatial data you can actually measure against. Hero Cloud claims to be through-the-lens reconstruction, meaning it extracts depth from the camera movement in the shot rather than relying on a separate photogrammetry set or a LiDAR scan.

The node targets a familiar production reality: you get a brief that needs measured 3D, but the the on-set… people…. said “Fix it in Post, heheeheeh” and did not capture the extra data for cases like set extension, CG integration, lighting reference, forensic analysis, architectural measurement, and archival footage where you cannot go back and reshoot reference. If the plate is all you have, Hero Cloud aims to make that plate pull double duty.

Hero Cloud works on plates PFTrack can track and solve, and works footage sources that include cinema cameras, drones, body cams, dash cams, action cameras, mirrorless and DSLR cameras, CCTV, and archival film. Basically, every type of video. Camera metadata can help but it is not required, and the Camera Sensor Database can cover common cameras automatically, with the Camera Solver able to estimate parameters when metadata is missing.

There is a catch, and it is the same catch as every method that tries to infer depth from motion: the shot needs parallax. Locked off shots or shots with very little movement can produce sparse or unreliable results. Most production footage, including handheld stationary shots, can still have enough movement to produce usable reconstruction, while deliberate camera motion tends to produce richer results.

A cleaner route into splat training

PFTrack 26.05.19 adds a Postshot export option that writes Hero Cloud point clouds and associated tracked cameras in COLMAP format. In other words, matchmove and reconstruction can stay in the same app up to the point where you hand off to splat training.

If you already live in the Gaussian Splatting world, COLMAP is a common interchange format and COLMAP itself has both GUI and command line workflows. The practical value here is that you do not have to glue together separate tracking, reconstruction, and export steps before you ever open Jawset Postshot. You track the plate, generate the cloud, export in COLMAP format, then train the splat.

This is an artist friendly route into a workflow that has historically leaned on command line tooling. That is a reasonable and recommendable goal, but it still deserves the usual pipeline skepticism: try it on a real shot and verify your coordinate frames, scaling, and camera consistency before you let it anywhere near a deliverable.

USD point clouds now carry more than positions

The same build extends USD export to support dense point clouds with per point normal and colour data, because a point cloud that only carries XYZ positions often becomes a dumb scaffold the moment it leaves its origin app. Normals and colour help downstream tools treat the cloud as more than a vague spatial fog, and can make it easier to use for measurement, placement, modelling reference, and lookdev decisions.

Improved integration with USD aware tools including Unreal Engine, Maya, and Houdini. It also claims (!) improved FBX export reliability for very large dense point clouds, aimed at users who previously hit issues exporting big clouds via FBX.

If your pipeline already standardises on USD, this is the kind of unglamorous export detail that can decide whether a new feature becomes a daily tool or a demo-only party trick. Still, test it on your actual asset scale and on your actual DCC import settings. Export support is never a substitute for shot-based validation, and you only find the sharp edges when production hits them, much like furniture at night can only be found with toes and shins.

Photo Mesh gets a single shot input path

Studio and Enterprise editions receive an update to Photo Mesh that enables it to consume Hero Cloud point clouds as input. That creates an internal single-shot path from tracked plate to dense cloud to surfaced textured mesh, without leaving PFTrack. The Pixel Farm describes Photo Mesh as the highest fidelity route when you have a planned photo set, while Hero Cloud covers the case where you only have the hero plate.

In practical terms, this puts a reconstruction option closer to matchmove, and it reduces the number of handoffs in the middle of a cmaera tracking job. Whether it replaces any external reconstruction tooling depends on what you already trust for meshing, texturing, and scale fidelity, but the internal integration is at least a coherent story: generate the cloud, feed it into Photo Mesh, then export into the rest of your pipeline.

The viewport tweak you will actually use

26.05.19 also adds an Active Camera viewport option in viewer windows, with a snap to active clip toggle. It sounds minor, and it is, which is why it will get used: quick visual verification against the source plate tends to catch mistakes faster than hunting through node outputs. The release notes pitch it as useful for checking Hero Cloud reconstruction, but the feature applies anywhere you want a fast way to see what the tracked camera saw.

Trial and licensing notes

Solo trial mode now enables full export functionality for seven days. The release notes describe this as enough time to take a plate from track through Hero Cloud and out to Postshot or USD, rather than being limited to evaluating the toolset without export.

For Studio, the release notes state it is available on rent to buy for 5 or 30 days, or as a perpetual purchase. The PFTrack Creator Program page states PFTrack Solo licenses sell for £699. Treat that as a price reference for Solo licensing, not a promise of regional pricing or bundles.

What this changes in real work

Hero Cloud aims to shift PFTrack from a pure matchmove and survey friendly reconstruction hub toward a single shot scene reconstruction step that can be used even when the onset data plan fell apart. The strongest parts of the release are not flashy, they are connective tissue: COLMAP export for Postshot, denser USD point cloud payloads, and a meshing path that stays inside the same graph.

It is also a reminder that the industry has started to treat point clouds as a first class asset, not only a temporary byproduct. When point clouds export with colour and normals and survive the trip into USD aware DCCs, they become reusable poitn based reference that can support layout, set extension decisions, and reconstruction driven previs.

https://www.pftrack.com/post/pftrack-build-26-05-19-is-now-live

https://www.pftrack.com/post/hero-cloud-tutorial

The post PFTrack gets Hero Cloud first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

For those who don’t know the tool: SplatTransform is a CLI and library that converts and edits Gaussian splats for pipelines that end up in SuperSplat or real-time runtimes. It ships as a package for Node.js and also runs in browser contexts.

What 2.0 actually changes

SplatTransform 2.0.0 adds a sparse voxel grid plus exterior fill for nav simplification. That pushes the tool beyond file shuffling and into spatial reasoning you can use for scene cleanup and interaction prep, especially in Gaussian Splatting workflows. Or, you know, previews.

Two new GPU-based processing actions land alongside the voxel pipeline: filterFloaters and filterCluster. They target unwanted bits that do not belong to coherent spatial clusters, and they run on the GPU path.

Floors, clusters, and tighter crops

2.0.0 adds floor fill, voxel-level cluster flood, and tighter voxel crops. The net effect is more structured scene operations, built on the voexl representation rather than only per-splat attributes. The release also refactors logging and progress reporting, and the CLI now reports peak memory usage in its final summary line. That is small, but it is the kind of small that saves time when you run batches overnight, or the whole studio is computing in timewarp.

Less native baggage, more inputs

The tool replaces meshoptimizer with a lossless coplanar-merge pass. That removes a native dependency from the meshing side of the pipeline and keeps the operation lossless by design. SplatTransform also accepts http and https URL inputs in the CLI, so you can feed remote assets directly without a manual download step. PlayCanvas also updates the dependency on supersplat-viewer as part of the 2.0.0 release work.

Formats, install, and the boring but important bits

The repo documents claim support for PLY, Compressed PLY, SOG, SPZ, SPLAT, KSPLAT and LCC, plus write support for PLY, Compressed PLY, SOG, SPZ, GLB, CSV, HTML Viewer, LOD and Voxel formats. So, everything we ever heard of as formats for splats. Install uses npm with a global command, and there is also a Docker backend guide for server-side use. Which we will be testing soon, and if you don’t hear from us again, something exploded.

The code is MIT-licensed and published as an open-source package, which usually means it’s free to use, but any production adoption still deserves a legal and pipeline check. New tools and innovations should always be tested before use in production, preferably on the worst shots and the messiest captures first.

https://github.com/playcanvas/splat-transform/releases/tag/v2.0.0

The post PlayCanvas updates SplatTransform to 2.0 first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

The problem NanoGS tackles

3D Gaussian splatting can look like magic right up until it starts behaving like a tax audit. Big captures mean millions of splats. Millions of splats mean constant sorting work, a lot of memory pressure, and a very real risk that the renderer spends time drawing things the camera never sees. NanoGS targets that exact pain: rendering large-scale Gaussian splatting files in real time, while handling video memory pressure, sorting huge splat counts efficiently, and drawing only what the screen needs.

The approach focuses on a set of optimisation techniques: Nanite-style LOD clusters, screen-space error LOD selection, splat compaction, a global accumulator, and GPU radix sort. The headline promise is straightforward: large-scale Gaussian splatting scenes should render efficiently, with an emphasis on keeping memory usage low. That is a marketing claim.

Supported engine versions and what you actually install

NanoGS lists support for Unreal Engine versions UE5.6 and UE5.7. The workflow stays inside the familiar plugin lane. You download the plugin from the releases page, place it into your project’s Plugins folder, then import a PLY file using the plugin import button. That import creates a Gaussian Splat Asset, which you can drag directly into a level for rendering.

There is also an asset action to enable or disable Nanite if needed. It is an unusual sentence to write about splats, but that is the point: the tool is trying to make splats behave like a native citizen in the realtime world.

The controls that decide whether it feels stable

The plugin exposes settings grouped by intent. They map cleanly to the trade-offs most realtime teams already juggle: quality, performance, and final look. On the quality side, SH Order adjusts spherical harmonics quality. On the performance side, Sort Every Nth Frame changes how often splats get sorted. Enable Frustum Culling turns frustum culling on or off. LOD Error Threshold sets how sensitive LOD cluster switching should be. On the rendering side, Opacity Scale adjusts splat opacity, and Splat Scale adjusts splat size.

These are not vanity sliders. Sorting frequency can be the difference between a stable camera move and a shimmering mess. Culling can be the difference between a scene that fits and a scene that explodes memory. LOD sensitivity can be the difference between crisp detail and visible popping. And opacity and scale are the levers that decide whether the splats read as a coherent surface or as a cloud of regret.

About one third into your first test session, you will probably notice the same thing most teams do: the scene can look fine at rest, then fall apart during motion if sorting gets too infrequent. That is when you stop guessing and start dialing.

Debug commands that make problems obvious

NanoGS includes console commands designed for inspection and stress testing. gs.ShowClusterBounds 1 enables a Nanite cluster preview, and setting it to 0 disables that preview. gs.DebugForceLODLevel can force rendering at a specific LOD cluster level for debugging, with values like 1, 2, 3, 4 and so on.

Then there is the one that will get used the most in anger: gs.MaxRenderBudget. It limits the maximum number of visible splats after culling to save memory. The default is no limit, using 0. You can set a maximum cap such as 3,000,000. When the cap is active, the culling starts from splats far from the camera.

If you are trying to ship anything on fixed hardware budgets, this kind of explicit cap is refreshing. It makes the failure mode predictable. It also makes it easier to have honest conversations with art and production, because the tool draws a hard line instead of pretending everything will fit if you just believe harder.

Best practice: do not feed it one giant splat brick

NanoGS calls out a best practice that will sound familiar to anyone who has ever fought streaming, culling, or draw lists. A single big chunk of gaussian splatting file does not help performance optimization. Splats outside the camera view cannot be culled effectively. And if everything stays in one chunk, all splats stay engaged in sorting all the time.

The recommended format is to slice a large splat file into smaller pieces. Examples given include individual props for cinematic scenes or tiles for geo-spatial data. The repository includes a simple tile slicer written in Python to support that workflow.

Proof of concept, not a miracle

NanoGS is described as a proof of concept, with a clear warning not to expect perfect results. Issues can be reported on the project page. That framing matters for production teams because it sets the correct expectation: you are evaluating a technique and an implementation, not adopting a finished commercial renderer. The good news is that the tool is built to be tested, with the kind of switches and debug views that make evaluation less mystical.

New tools and innovations should always be tested before use in production, especially when they touch core rendering paths and memory budgets.

Release, license, and what it costs

NanoGS is released under the MIT license.

The repository lists v1.0.1 as the latest release, dated March 20, 2026.

The plugin is free.

https://github.com/TimChen1383/NanoGaussianSplatting

The post NanoGaussianSplats for UE5 first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

For those who don’t know the event: FMX is where post, VFX, and realtime folks compare notes, spot new workflows, and meet toolmakers from Foundry to AYON .

The 30th edition runs on site May 5 to May 7, with Online and On Campus on May 8, and On Demand from May 9 to June 9. Early Bird ticket options run until March 20. Tickets are sold through the ticket shop, and the full schedule is published as an online program schedule.

This year’s reveal is a broad spread across VFX, animation, and digital media creation, with more updates promised as additional speakers and topics get added. The headline message is simple: the program is live now, and it is designed to be browsed like a buffet. Get your preferred morsels here :)

RENO turns a studio inward

A newly confirmed session in the Shorts track focuses on RENO, an original sci fi short used as a live testbed for virtual production and machine learning workflows. The session names Tav Flett, Rob Hifle, James Pollock, and Paul Silcox, and describes how a traditionally post-driven VFX studio re-engineered its pipeline for a newly opened virtual production stage. It also calls out Gaussian splat scanning, automated rotoscoping, and marker removal as tools for streamlining production.

Generative video, now with shot notes

The hype cycle is almost over. The bubble tends to pop the moment somebody asks where this stuff fits into an actual production pipeline. That is the useful angle of Moonvalleys‘ session in the “The Road ahead” track. . Instead of treating generative video as a party trick, Ben Lock looks at what happens when studios try to use it for real: control, continuity, versioning, integration, governance, and all the other small details that usually kill shiny ideas on contact.

For VFX and animation teams, that is the point. Not more halfbaked sizzle reels, but a sober look at where generative video already helps, where it still breaks, which parts of the panic were never grounded in production reality, and when studios may begin hiring artists again once practical applications, and their many many limits, are easier to assess.

Predator: Badlands, built as a handoff

A joint presentation in the VFX for Features track names Laurens Ehrmann from The Yard VFX and Dominik Zimmerle from TRIXTER discussing work on Predator: Badlands. Shared assets, shared look development, and constant creative exchange across studios to ensure continuity while elevating environments, characters, creatures, and effects, plus a deep dive into selected sequences and the creative and technical challenges behind them.

And there is more to come! Much more!

https://fmx.de/en/program/program-2026/overview

The post FMX 2026 program goes live first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

Placement in the Ecosystem: glTF from the Khronos Group is a lightweight asset format for realtime engines, viewers, and DCC export, commonly used between tools rather than inside them, and now it wants splats too. Visit their site here.

A careful step into splat territory

The Khronos Group has published a release candidate for a new glTF extension that adds support for Gaussian splatting. The proposal, named KHR_gaussian_splatting, defines how 3D Gaussian splats can be represented inside glTF 2.0 assets using a common, vendor-neutral schema.

Gaussian splatting is a scene representation technique where geometry is described as large numbers of oriented 3D Gaussians rather than polygon meshes. Each splat typically carries position, scale, rotation, colour and opacity. Rendering relies on blending these splats in screen space to approximate surfaces and volumetric detail. The approach is increasingly used in photogrammetry-derived captures and radiance field research, particularly where dense detail and rapid rendering are required.

Until now, Gaussian splats have circulated in a growing number of ad hoc file formats. Khronos is explicitly positioning this extension to prevent early fragmentation by defining a baseline encoding before incompatible variants become entrenched.

What the extension actually defines

KHR_gaussian_splatting introduces a standardised way to store splat data in glTF without redefining the rendering algorithm. The extension specifies how splats are described as primitives and which attributes are required or optional. These include spatial transforms and per-splat appearance data, mapped onto existing glTF data structures where possible.

The proposal does not mandate a specific splatting technique or optimisation strategy. Khronos describes the design as algorithm-agnostic, with the goal of supporting multiple implementations rather than locking vendors into a single approach. This restraint is notable given how quickly splat rendering methods are evolving.

Viewers that do not support the extension are expected to fall back gracefully, typically treating the data as a point cloud. This preserves basic visibility rather than failing outright, which aligns with glTF’s long-standing emphasis on predictable degradation.

Why Khronos is getting involved now

Gaussian splatting has moved from research into production-adjacent tools at a speed that has made standardisation awkward but necessary. Khronos frames this release candidate as a response to real usage rather than speculative future demand. The organisation is inviting feedback from developers and vendors before final ratification, signalling that the format is still open to adjustment.

The work builds on earlier collaboration between Khronos and geospatial and platform vendors exploring splat-based representations for large-scale scenes. While those efforts are not detailed in the extension itself, the context suggests a desire to make splats a first-class citizen alongside meshes, animations and textures in glTF.

glTF is often described by Khronos as a transmission format rather than a working format. Adding splats does not change that positioning, but it does acknowledge that realtime pipelines are absorbing data types that were previously confined to research or proprietary viewers.

Interoperability over novelty

There are no promises of faster rendering, smaller files, or better fidelity. Instead, the emphasis is on interoperability and predictability. This is a conservative but welcome stance for a standard that already underpins web viewers, game engines and real-time visualisation tools.

Compression is not part of the core proposal. Khronos notes that companion extensions may be introduced to improve storage efficiency, but they are not part of the current release candidate. No compression scheme is ratified as part of KHR_gaussian_splatting yet.

This separation is sensible. It allows the base representation to stabilise before additional complexity is layered on top, and avoids repeating past mistakes where format and codec decisions were entangled too early.

A positive but cautious signal

As proposed, KHR_gaussian_splatting does not attempt to redefine how Gaussian splats are generated or rendered. It simply gives them a place to live in a widely adopted container format. That restraint is arguably its strongest feature.

For production teams experimenting with splat-based capture or visualisation, a common interchange format reduces risk. For tool developers, it provides a reference target rather than another reverse-engineered file. For Khronos, it reinforces glTF’s role as a neutral meeting point rather than a battleground of features. As always, new formats and extensions should be tested thoroughly before being relied on in production environments.

// Khronos Group press release on glTF Gaussian Splatting
// https://www.khronos.org/news/press/gltf-gaussian-splatting-press-release

The post glTF learns to splat without losing the plot first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

The Visual Effects Society (VES) has announced that its flagship reference, the VES Handbook of Visual Effects, reaches its Fourth Edition on December 18, 2025. The book now brings together 95 visual effects experts, including 30 new contributors, from across film, television, games, virtual production, and emerging media.

The new edition expands its scope to modern practices. Among the additions are chapters covering the use of real time game engines for virtual production, previs, motion capture and animation. The handbook addresses current topics such as AI applications in VFX, plus methods for NeRFs (Neural Radiance Fields) and 3D Gaussian splatting, aiming to reflect recent workflow evolutions.

Existing core areas have also been updated. These include previs (pre visualisation), AR and VR filmmaking, colour management with attention to ACES workflows, digital intermediate processing, camera workflow, matte painting, compositing, and more.

Early peek event: live from the makers

Shortly before the book’s release, VES will host a global webinar on December 16, 2025. During that event, editors Jeffrey Okun, Susan Zwerman and Susan Thurmond O’Neal will discuss the new edition in conversation with industry experts, including virtual production TD from Industrial Light & Magic Nathan Camp, creative technology veteran Phil Galler, Chief Innovation Officer at 180 Studios, and VFX supervisor Lucien Harriot, President of Mechanism Digital. Attendees will get a preview of key workflows and techniques covered in the handbook and have the opportunity to ask questions to the contributors.

The editors: who’s behind the Fourth Edition

Jeffrey A. Okun is an award winning Visual Effects Supervisor, a Fellow of the Visual Effects Society, and a member of the Academy of Motion Picture Arts and Sciences, the American Society of Cinematographers, and the Television Academy. He has served as chair of VES, and previously as first vice chair and chair of the Los Angeles section.

Susan Zwerman is a long time VES producer with over 25 years of experience in visual effects production. She is also a respected seminar leader and author. Zwerman is a member of the Academy, the Producers Guild of America, and the Directors Guild of America, and holds the status of VES Fellow.

Susan Thurmond O’Neal joined VES in 1997 and has held multiple leadership roles over the years. She has served on the global Board of Directors, including as Treasurer in 2016, 2nd Vice Chair in 2022–2023 and 1st Vice Chair in 2024–2025. She has chaired the global Education Committee, currently leads the Membership Committee, and in 2019 received the VES Founders Award for service to the society.

Why this matters for VFX, CG and post production pros

As pipelines increasingly adopt real time engines, AI assisted tools, and immersive production formats such as virtual production, AR, VR and domes, this new edition of the VES Handbook provides a broad and current reference. The inclusion of NeRFs and 3D Gaussian splatting signals that VES acknowledges emerging volumetric and point cloud based workflows as significant enough to warrant formal documentation.

For those working in compositing, colour grading, camera and layout workflow, the updated coverage of core disciplines, re verified and aligned with contemporary practices, provides a stable reference against a rapidly shifting technological backdrop.

The breadth of the handbook, from budgeting and shot breakdowns to high end compositing and immersive media workflows, suggests it aims to be relevant across departments: production, CG, compositing, virtual production teams, colour, and more.

Get a sneak Peak into the Summary here on Google Books:

Finally, the new standard twist

With the Fourth Edition, the VES Handbook doubles down on being the industry standard reference. By combining decades long experience with up to date coverage of real time engines, virtual production, AI, volumetric rendering and immersive media, this edition positions itself as the reference manual to return to when technologies or pipelines evolve. It may not promise radical revelations everywhere, but it maps out the best practices terrain for the current generation of visual effects work. For any VFX professional aiming to stay grounded and informed, this looks like the old new “go-to” compendium.

The post VES Handbook of Visual Effects, Fourth Edition, coming soon first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

The browser-based editor SuperSplat from PlayCanvas Ltd. is now on version 2.14, bringing key video rendering features that make it far more viable for production work. These updates are grounded in official release notes and the user-manual. Marketing claims are noted where used. As always, test in your pipeline before committing to live work.

https://twitter.com/willeastcott/status/1990445129703043376

What SuperSplat is

SuperSplat is a free, open-source editor for 3D Gaussian splats (“splat” meaning a billboarded Gaussian sphere used to reconstruct point clouds, often for realtime visualisation). It runs fully in the browser (no install), supports import/export of point-cloud/splat formats, offers visual editing, optimisation and publication workflows. The tool is part of PlayCanvas’s Gaussian splatting documentation.

What’s new in version 2.14

According to the GitHub release notes for version 2.14.0: Support has been added for WebM (AV1 and VP9) encoding. MOV and MKV video encode support has been added. The timeline frame-range for video encoding has been exposed, giving you control of which frames to render. Additional framerate options for video export are now available. The video format selection has been separated from codec selection (so you can e.g., pick format and codec independently). A small shader fix for the latest engine was also included. These changes clearly move the tool from “interactive preview” to “output ready” for video fly-throughs, which is notable for post-production, VFX and realtime graphics workflows.

Recommended usage in production

Consider incorporating version 2.14 of SuperSplat into your pipeline when you have point-cloud/splat data that needs visualisation or fly-through output. Use the timeline frame-range and framerate controls to standardise output (e.g., 24 fps, 30 fps). Choose container formats that match your downstream: MOV/MKV for mastering, WebM for web previews. Clean up splat data (remove stray “floaters”, background noise, optimise count) prior to export to reduce overdraw and performance cost. Keep an eye on the visual fidelity vs fragment cost trade-off when exporting to high resolution video.

Note: As with any new tool or feature in a production pipeline, you should test SuperSplat 2.14 end-to-end before deploying to live work.

The post SuperSplat 2.14 adds full video export first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

With its second major release, Atlux λ 2 from Vinzi upgrades Unreal Engine from a lighting and rendering sandbox to something more unexpected: a fully simulated 3D scanning studio. The plugin, previously known as MetaShoot, already recreated physical photography stages within Unreal. Now, its new Capture tab extends this idea to the world of 3D data acquisition. Artists can distribute virtual DSLR cameras around a subject using the Camera Volume, tweak focal length, aperture, and shutter speed, and generate synthetic scan data without leaving the engine.

That data can be exported as standard PLY point clouds or COLMAP datasets, complete with camera positions, ready for reconstruction or use in Gaussian Splatting workflows. In effect, Atlux λ 2 allows Unreal scenes to feed directly into neural or point-based renderers—turning virtual scenes into training data or splat clouds.

Lighting: modernised for Lumen

The original Atlux lights were tuned for Unreal Engine 4’s raytracer. Version 2 retunes everything for Lumen, Unreal’s realtime global illumination system. Point and Directional light types have been added, Spotlight behaviour refined, and new presets included for standard studio setups.

A new procedural gobo system projects colour and shadow patterns across surfaces or acts as a softbox diffuser. The system remains parametric, letting artists adjust projection density, sharpness, and tint directly in the viewport.

Props, particles, and calibration

Atlux λ 2 also expands the library of assets available inside its Prop tab. New podiums, curtains, and cyclorama-style backgrounds can be dropped in with a click. Environmental effects such as light trails, particle bursts, fog, and god rays simulate studio atmosphere or help block out cinematic moods.

For technical accuracy, Vinzi adds reference materials and calibration props: chrome and grey balls, Macbeth-style colour charts, and surface decals such as fingerprints or scratches for realism checks.

Cameras and animation

Camera motion receives a complete overhaul via the new Action tab. Beyond static turntables and rail rigs, the update introduces floating, breathing, and handheld motions—useful for emulating product demo or cinematic coverage. Object animation presets now include oscillation and platform rotations, making it easier to preview assets dynamically without separate sequencer setups.

Stylisation and colour management

The Post-Process tab offers a surprising detour into non-photorealistic rendering. New materials simulate hatching, blueprints, oil painting, or even LEGO-style mosaic looks. While arguably niche, these presets demonstrate Atlux’s procedural layering system. Colour workflow has also matured. Artists can now load custom LUTs in .cube format, and use OCIO configuration directly from the plugin. This aligns Atlux with modern VFX colour management pipelines, especially for teams delivering look-dev stills or promotional imagery.

Randomisation and stability

Version 2 replaces the former AI Randomizer with Afflatus, a fully controllable randomisation engine. It can vary lights, props, or cameras individually, making batch renders or creative explorations more predictable. Vinzi has also addressed long-standing stability issues, notably a crash that occurred when cancelling renders mid-process. The UI has been reorganised and cleaned up, reflecting Atlux’s shift from experimental plugin to mature production tool.

Pricing and compatibility

Atlux λ 2 supports Unreal Engine 5.4 and later, running exclusively on Windows. It’s a free update for existing licence holders. A new perpetual licence costs $495 for individuals, while subscription users now pay $59.50 per month. Studios above $5 million annual revenue pay $990.50 or $109.50 per month.

Why a virtual scanner makes sense

Atlux’s decision to simulate photogrammetry inside Unreal feels both logical and overdue. Real-world 3D scanning is constrained by lighting, calibration, and physical capture limits. Unreal’s virtual environment removes all of these.

Because Atlux generates PLY or COLMAP data from synthetic cameras, users can prototype real scanning setups virtually—testing camera count, exposure, or spacing before investing in physical rigs. This also means Unreal can now output the same kind of data that Gaussian Splatting software ingests.

For visual effects, this unites two previously separate domains: engine-driven rendering and point-based reconstruction. Artists working on virtual production stages or motion-capture shoots could simulate scans of props or performers within their live Unreal environments, immediately feeding that synthetic data into splatting workflows or neural rendering experiments.

The result is a hybrid form of asset creation – half DCC, half dataset generation – without needing to step outside the engine.

The Gaussian bridge

Gaussian Splatting relies on point-based representations with density and colour information. Until now, these were usually reconstructed from photogrammetry or radiance field capture. Atlux’s virtual scan export removes the dependency on real footage: it produces consistent, noise-free data that can serve as either training material or direct renderable geometry.

It’s a small technical leap with large implications. For game cinematics, motion capture pre-visualisation, or virtual set planning, Gaussian splats generated in-engine would integrate seamlessly with existing Unreal lighting and compositing systems. Scene layout, lighting, and reconstruction could all occur within a single runtime context.

It raises a simple question: why did no one integrate a virtual scanner sooner? The pieces like Unreal, COLMAP, splat renderers, have existed for years. Atlux is simply the first to package them coherently inside the engine.

Practical caveats

Synthetic scanning does have limits. The generated data mirrors only what exists in the scene: no micro-surface irregularities, no photographic noise. For production pipelines requiring physical accuracy, real scans will still dominate.

Domain bias also matters. Gaussian reconstruction tools trained on real images may respond differently to perfect, noise-free synthetic data. Colour, specularity, and occlusion gaps might require deliberate simulation of imperfections to ensure realism.

Performance remains a concern. Simulating tens or hundreds of high-resolution cameras for each capture can tax GPU memory and I/O bandwidth. Unreal users will need to balance fidelity against throughput, especially when batch-generating scans.

A maturing plugin ecosystem

With Atlux λ 2, Vinzi seems to position Unreal less as a render target and more as a production laboratory—a controllable environment for both image generation and 3D data experimentation. It mirrors a broader shift in the industry: tools that blur the boundaries between DCC, realtime, and machine-learning workflows. As engines mature, the distinction between rendering, scanning, and simulation narrows.

Artists and technical directors should, as always, test the plugin in context before committing to production use. But Atlux λ 2’s ability to create synthetic scans directly inside Unreal marks a thoughtful and technically literate step toward unified virtual production pipelines.

Atlux Official Website

Atlux λ for Unreal 5.4 – Medium Announcement

The post Atlux λ 2 brings 3D scanning inside Unreal first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

USD support: finally opening the door

The headline addition in Vantage 3.0 beta is initial support for USD. Artists can now load .usd, .usda, .usdc, and .usdz files directly into Vantage, making it possible to preview both static and animated assets without going through V-Ray export first. Geometry types already supported include static meshes, animated transforms, and even fully deforming animated meshes. Particle systems can also be imported via USD.

Materials and attributes are carried across as well. USD Preview Surface materials are recognised alongside V-Ray materials and textures, meaning both native USD shading definitions and V-Ray-specific lookdev data can be rendered. V-Ray displacement, shader animation, and light parameter animation are all supported. Even user-defined attributes, such as those accessed via VRayUserScalar or VRayUserColor textures, are included. Importantly, material overrides assigned in USD also function in Vantage.

USD support extends to MaterialX as well, which ties directly into another first-time feature in this release. According to the Chaos Vantage 3.0 documentation, USD versions 0.10.0 and later are supported.

MaterialX and a new in-app material editor

Alongside USD, Vantage 3.0 introduces initial support for MaterialX materials. This gives artists a pathway to render and exchange open-standard shader graphs, not only proprietary V-Ray materials. For the first time, Vantage also includes a built-in material editor. Users can now create and edit materials directly within the application. However, only materials authored inside Vantage can be edited this way—imported materials remain locked. This ensures that Vantage can function not only as a passive viewer but as a limited authoring environment for shader experimentation. MaterialX support, combined with USD, establishes Vantage as a renderer that can sit earlier in asset pipelines and does not require full dependence on Chaos’ proprietary shading ecosystem.

Gaussian Splats: point clouds meet ray tracing

The second core addition is support for Gaussian Splatting. Gaussian Splats can now be rendered from .vrscene files and through Live Link connections. This allows scanned environments or point-based geometry representations to be displayed in real-time alongside traditional geometry.

The Gaussian Splats are ray traced inside Vantage, not rasterised. This ensures correct behaviour in lighting, shadows, reflections, and refractions, unlike typical point-cloud preview tools. There is no support for directly importing splat datasets in formats such as .ply or .xyz at this time; they must be embedded in .vrscene assets or streamed through Live Link. The Chaos documentation notes this is a first implementation, which may expand in future versions. For now, it already offers a practical path to bring photogrammetry or neural scene representations into GPU-accelerated previews.

Volume rendering: smoke and fire enter Vantage

Chaos has added initial support for volumetric rendering. Users can now load caches from Chaos Phoenix in .aur format as well as V-Ray volumetric grids in .vdb. Vantage provides a new toolbar option, Create Volume Grid, to insert and manage volumes within the viewport. Shading options remain limited in this beta stage, but the feature allows real-time preview of effects such as smoke, fire, or clouds directly in the renderer. This update expands Vantage’s potential role for lookdev on FX shots, where volumes often play a key visual role.

Scatter now supports clustering

Scatter has been expanded with support for object clustering. Users can now group scatter instances together, improving control and natural distribution patterns in instanced geometry setups. However, instance layer painting and colour map controls are still not supported. The clustering update improves scatter’s usability for layout, but without the full range of painting and map-driven workflows, it remains less advanced than DCC-based scatter systems.

Texturing improvements across engines

Textures receive several technical updates. The MultiSubTex map now honours probability parameters, enabling more precise randomisation. V-Ray’s VRay2SidedTex is supported, allowing two-sided shading definitions to display correctly. Projection mapping support has expanded too: Maya projection textures are now usable, and Cinema 4D projection types such as screen and frontal mapping are supported. These additions allow Vantage to correctly preview a broader variety of materials originating from common DCCs.

Camera tracking for virtual production

A significant addition for on-set workflows is support for external camera tracking protocols. Vantage 3.0 beta can now receive tracking data from LONET 2, EZtrack TCD, FreeD, and stYpe HF. This enables synchronised camera motion between real-world tracked cameras and Vantage’s viewport. In virtual production contexts, this allows Vantage to serve as a real-time renderer connected directly to camera hardware, expanding its role beyond pre-rendered lookdev.

New view mode for selection

Vantage 3.0 beta adds a Selection Mask view mode. When enabled, selected objects render in white while everything else renders black. This assists in identifying, isolating, or troubleshooting geometry within heavy production scenes.

User interface: palette and workflow refinements

The application interface has been updated with a new colour palette and overall redesign. The home screen has been reworked, and side panels are now resizable for better layout flexibility. Camera objects can be edited directly inside the Details panel under the Object tab. However, changes are not reflected in the viewport until the camera is re-selected from the Camera Lister. This quirk is noted in the official documentation. The old “Render setup” has been replaced by a Render mode, located at the top-right of the toolbar, simplifying access to key rendering controls.

Render elements: name remapping

Vantage 3.0 introduces an environment variable for remapping render element names. The variable VRDX_REMAP_ELEMENT_NAMES allows artists to redefine names on output, ensuring downstream compositing or pipeline systems can process them with consistent labels. This function provides a simple but effective layer of flexibility for facilities that rely on strict naming conventions.

Fixes across file handling, textures, displacement and lights

Several targeted bug fixes are included in the beta. File handling has been stabilised: crashes caused by .vantage files referencing deleted animated cameras in linked .vrscene assets are resolved. In textures, rendering errors with large HDRI maps on Blackwell GPUs have been corrected. Problems with animated textures when DLSS denoising was enabled are fixed. Blender scenes with multiple UV channels now map correctly.

Displacement issues are addressed too. Per-triangle tessellation displacement no longer causes crashes when applied to .vrmesh assets containing multiple geometry objects. Lighting fixes include corrected rendering for light meshes built from multi-geometry objects, and better handling of prototype flags in light instances. Procedural objects such as Chaos Scatter now properly respect light linking, and previously observed black artifacts on directly visible area lights have been eliminated.

Final Notes on This Beta Release

The Chaos Vantage 3.0 beta brings long-requested interoperability with USD and MaterialX, the experimental addition of Gaussian Splats, first-pass volume rendering, expanded scatter tools, and multiple workflow refinements. For real-time lookdev and layout tasks, it is a substantial update.

However, much of the support is explicitly marked as initial. Both USD and MaterialX import are functional but not comprehensive. Gaussian Splats require .vrscene or Live Link sources rather than standalone datasets. Scatter lacks painting and colour map controls. Volumes remain limited in shading controls.

As with any beta, these features should be tested in controlled pipeline environments before being relied upon for production. That is the point of a Beta Version!

Apply HERE if you want to take part.

The post Chaos Vantage 3.0 beta lands with USD, MaterialX and Gaussian Splats first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

4Dviews has announced a major upgrade to its HOLOSYS+ volumetric capture system: native support for 4Dviews Gaussian Splatting as an export option alongside the traditional mesh‑and‑texture output. This marks the first time a volumetric capture system offers both formats natively, giving pipeline teams technical and creative flexibility without changing capture workflows. The company emphasizes that Gaussian Splatting is another tool to fuel narrative control—not constrain it. This approach underscores the modular and scalable fidelity design that underlies ongoing R&D at 4Dviews. 4Dviews will demonstrate Gaussian Splatting support at SIGGRAPH 2025 in Vancouver from August 10 to 14 at Booth 417.

4Dviews Gaussian: Splat or Mesh, You Decide

The new dual‑export pipeline enables creators to choose between standard mesh‑and‑texture assets or Gaussian Splatting output from HOLOSYS+ captures. Gaussian Splatting excels at rendering complex material characteristics, such as hair, fur, reflective or translucent surfaces, and fine props, while maintaining compatibility with existing HOLOSYS data pipelines.

4Dviews states that this integration is engineered for stability and visual fidelity under challenging capture conditions. The system supports dynamic performance capture suitable for low‑bandwidth streaming or high‑fidelity offline rendering while preserving backwards compatibility with all previously captured volumetric assets.

What It Means for Teams

Production teams now have access to two volumetric rendering paths: traditional mesh workflows for compatibility with existing VFX and editing pipelines, and splatting workflows for improved handling of materials and fine details. This dual choice may reduce compromises during capture and postproduction for assets with complex visual traits.

Notably, although Gaussian Splatting enables enhanced surface detail, each production should test the format choice under actual production conditions before integrating into deliverables.

The post Dual Export: 4Dviews Adds Gaussian Splatting to HOLOSYS+ first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

Anyone working with Gaussian Splatting in Houdini can thank David Rhodes and Ruben Diaz for the kind of open-source toolkit that doesn’t just patch holes, but gives artists an entire toolbox. Now under the CG Nomads banner, GSOPs 2.6 lands as a free, AGPL‑licensed Houdini Digital Asset package—ready to import, edit, relight, mesh, animate, and export both 2D and 3D Gaussian Splats, all from inside Houdini 20.5.

Built for Houdini Artists, Not Marketers

Installation isn’t a mystery hunt: clone the develop branch, open the bundled installer HIP file, and (optionally) grab sample data with a checkbox. The full suite is compatible with Houdini 20.5 on Windows and macOS; Linux gets no official love, at least for now. Documentation and a detailed wiki are live on GitHub, and yes, it really is free—unless you want Early Access, which adds some exclusive toys for $10/month.

Not Just 3D—2D Splats Arrive, Too

While GSOPs has supported 3DGS (.3DGS) files for a while, v2.6 brings in support for 2DGS (.2DGS) files as well. Artists can load, render, and edit both 2D and 3D Gaussian Splat models right in Houdini’s viewport, with real-time GPU rendering and full attribute support. No more conversions or third-party hops just to view your splats.

Viewport Renderer and a Node Roster for Everything

The GSOPs SOP suite is far from a one-trick import node. Here’s what you actually get:

• Importers for .ply, .2DGS, .3DGS models, converting them to native Houdini point clouds, including spherical harmonics (SH) for view-dependent rendering.
• Viewport Source SOP: Real-time GPU-based splat rendering.
• Alignment and Transformation: Tools like gaussian_splats_align_by_points and gaussian_splats_transform allow full reposition, rotation, and per-point scale, with SH support baked in.
• Clean-Up Nodes: Use gaussian_splats_dbscan to remove noise or isolate regions with gaussian_splats_crop.
• Attribute & Color Control: Adjust SH coefficients, tweak attributes, or drop a Hald CLUT for color corrections.
• Deformation & Animation: gaussian_splats_deform enables mesh-driven deformation, and a set of “toys” includes advect, font, and jellify (Vellum soft-body splats).
• Mesh Generation: Convert splats to coarse meshes using nodes like vdb_from_gaussian_splats, or build sparse polygon graphs for further edits and downstream workflows.
• Synthetic Data Output: The gaussian_splats_generate_training_data SOP renders out training datasets (PNGs, COLMAP cameras) and is compatible with Karma and third-party renderers.
• Relighting and IBL: Relight splats with image-based lighting (gaussian_splats_relight_ibl), exploiting SH for dynamic effects.
• Exporters: Dump splatted models back to disk with full point and SH data for round-trip pipelines or external use.

Early Access: Karma & Solaris—Now with USD Integration

Those who support CG Nomads on Patreon get a little extra: USD integration and Solaris-native LOP nodes, including the “Gaussian Splats Import LOP.” This node injects splats into the Solaris USD pipeline, baking SH into Houdini’s v@Cd color attribute. Result? Real splat relighting, shadows, and reflections in the Karma renderer—and direct compatibility with USD-based lookdev and lighting workflows. Early Access also unlocks premium demo scenes.

Motion, Synthetic Data, and Feature Analysis

GSOPs isn’t just about static scans. It supports playback and editing of motion sequences—one splat file per frame—for workflows in Unity, Postshot, or SuperSplat. The toolkit also generates synthetic data for AI training or procedural content, and includes nodes for statistical analysis and data cleanup, giving artists deeper insight into their splat clouds.

Community, Support, and Licensing

GSOPs is AGPL-licensed—so it’s open for most studios and indie teams, with custom licensing for larger commercial studios if required. Community feedback and bug tracking are handled via GitHub, with further community showcases on LinkedIn and video reels linked in the repository. GSOPs recently placed third in the SideFX Houdini H20 Tech Art Challenge, earning a nod from the developer crowd.

What’s Missing? Official Linux Support and Caution

No official Linux support yet—Windows and macOS only. Documentation is extensive, but as with any new rendering pipeline, test every feature in your own environment before deploying in production. Real-time viewport speed depends on GPU horsepower and splat count; massive scenes may challenge Houdini’s interactive preview.

Closing Note for Cautious Professionals

GSOPs 2.6 is the most comprehensive Gaussian Splatting toolkit yet for Houdini, opening the full edit–relight–mesh pipeline for both static and animated point-based volumetric models. Karma and Solaris USD integration makes it genuinely production-minded, not just a tech demo. But, as always: test before you trust, especially if your deadline depends on it.

Cited Sources

// GSOPs official docs, feature list, node breakdown, licensing, installation // GSOPs – cgnomads/GSOPs GitHub
// GSOPs node documentation // GSOPs Wiki – github-wiki-see
// Karma/Solaris features, demo scenes // LinkedIn post by Ruben Diaz

The post GSOPs 2.6: The Swiss Army Knife for Gaussian Splatting in Houdini first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

Unigine has released SDK version 2.20, and if you think “Gaussian Splatting” sounds messy, you’re right—but it’s also the latest tool for high-fidelity point-cloud rendering, now available for simulation and engineering users. But Unigine 2.20 doesn’t stop there: the update delivers a raft of rendering, workflow, performance, and VR/XR enhancements, all thoroughly documented in Unigine’s official release notes and developer log.

Gaussian Splatting: For the Brave and the Enterprise

Users of the Engineering and Sim SDKs (that’s the not-free tiers) can now import and render .ply point cloud data as 3D Gaussian splats—an experimental plugin that comes with its own sample scene and setup tools. If your idea of fun is high-density reality capture, now you can splat with the best. Blender, Unreal, Unity, and V-Ray have all jumped on this bus; Unigine keeps pace for serious enterprise workflows.

Visual UI Toolkit: Drag, Drop, Repeat

Tired of gluing together UI code in the dark? Unigine 2.20 introduces a visual UI Designer and Runtime Editor: assemble widgets in a WYSIWYG environment, snap to grid, undo your mistakes, all with live preview and JSON-based layouts. The new UI system is fully scriptable, supports runtime debugging, and bundles a widget set that covers everything from buttons and sliders to tables and containers. No more hunting for missing pixels.

Rendering Pipeline: Smoother, Sharper, Faster

On the rendering front, Unigine 2.20 brings Vulkan-based DLSS support and updates AMD FSR upscaling to 3.1.3 on both Windows and Linux. Octahedral impostors (think: sharper distant billboards), improved voxel probes for subsurface and scattering, new global illumination controls, and sharper image effects all land in this release. The Agility SDK integration lets developers access the latest DX12 features even on Windows 10—if you still have to.

Multithreaded Shadows and Lighting: Shadows in a Hurry

DirectX 12 users will see multithreaded shadow rendering—up to 3x CPU performance gain in dense lighting scenarios. Shadows, projectors, and omni lights all move to mesh-based rendering for smarter culling and throughput. If you spend your day staring at scene graph bottlenecks, it’s time to reboot.

Performance and Memory: The Never-Ending Quest

Mesh compression now brings smaller binary clusters, with load/save times improved by up to 20%. The memory allocator has been rebuilt for lower VRAM spikes, and terrain clutter/grass intersection is now ten times faster. RAM/VRAM footprints shrink across particles, billboards, and visualizers. Startup and global resource loading are also noticeably speedier.

VR/XR: Real-World, Real-Time, Ready

XR users get eye-tracking, dynamic foveated rendering, and mixed-reality passthrough for Meta Quest 3 and Varjo XR-3. Hand tracking, per-eye camera offsets, better HMD emulation, and adaptive exposure mean VR workflows aren’t left in the 2020s. The input system, device compatibility, and sample scenes all see updates. Passthrough mode for MR is finally stable enough for the show floor (your mileage may vary).

Editor, API, and Scripting: Fixes, Tweaks, and More Power

Unigine Editor gets a search overhaul, property panel speedup (5x faster), shadow baking that no longer tests your patience (up to 17x), improved window layout persistence, and better graph tools. The C# component workflow now supports interfaces, abstract classes, named access, and external libraries—good news for pipeline maintainers. Expect dozens of bug fixes, small usability upgrades, and a more stable scripting environment overall.

Terrain, Samples, Docs, and the Rest

Sandworm (terrain tool) gets memory and speed improvements, SpiderVision picks up line rendering, and the IG module adds CIGI packet support. C++ and C# sample code is now up-to-date on GitHub. Documentation, VR/MR/hand-tracking guides, API migration notes, and new video HowTos round out the update.

Platform and Pricing

Unigine 2.20 runs on Windows 10+ and Linux 4.19+, 64-bit only. The Community (Free) SDK remains free for non-commercial or sub-$100K projects; Community Pro costs $1,500/year. Engineering and Sim SDK pricing is available on request.

All innovations should be tested in non-critical environments before rolling into production—bleeding edge, after all, is still called bleeding for a reason. UNIGINE Devlog: 2.20 Release

---

The post Unigine 2.20: Now with More Splat and a Whole Lot Else first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

In the ever-evolving world of visual effects, staying ahead means embracing tools that blend innovation with practicality. Enter Gaussian Splatting for Nuke, a plugin developed by Irrealix, designed to import, manipulate, and render 3D Gaussian Splatting (.ply) files within Nuke’s compositing environment. This plugin promises real-time performance powered by GPU acceleration, aiming to streamline workflows for VFX professionals.

What Is Gaussian Splatting?

Gaussian Splatting is a rasterization technique that enables real-time rendering of photorealistic scenes from minimal image samples. By approximating point clouds with Gaussian functions, it reduces computational load while maintaining visual fidelity—a boon for artists seeking efficiency without compromising quality.

Key Features

The plugin offers a suite of functionalities tailored for seamless integration into Nuke:

• File Handling: Import .ply files, including those compressed with SuperSplat, and combine up to 10 models within a single scene.
• Cropping & Alignment: Crop scenes using spherical or box shapes, or the Y Plane. Align models to the scene’s center and orient them horizontally, with orthographic projection cameras available for precise alignment.
• Color & Effects: Colorize splats with ramps using spherical or box shapes. Animate splat scale and opacity with ramps for dynamic reveals. Apply noise to distort models, and perform individual color correction for each model to adjust temperature and brightness.
• Real-Time GPU Acceleration: Manipulate models in real-time, thanks to GPU optimization, ensuring a smooth workflow.
• Depth & Rendering: Render depth passes for 3D composition and export scenes seamlessly for further processing or sharing.

These features are designed to enhance the efficiency and creative possibilities of integrating 3D data into advanced compositing workflows.

Workflow Integration

By enabling the import and manipulation of point cloud data directly within Nuke, Gaussian Splatting for Nuke eliminates the need for external rendering engines. This integration allows artists to interact with 3D data in real-time, facilitating immediate feedback during compositing tasks—a significant advantage in high-pressure post-production environments.

Availability and Pricing

Gaussian Splatting for Nuke is available for purchase at $49.00. For more information and to acquire the plugin, visit the aescripts website.

Final Thoughts

While Gaussian Splatting for Nuke introduces promising capabilities for real-time 3D rendering within compositing workflows, it’s advisable to evaluate its performance within your specific production environment. As with any new tool, ensuring compatibility and stability is key to maintaining the seamless operation of your VFX pipeline.

The post Nuke Gets Splatted: Gaussian Splatting Plugin Debuts first appeared on DIGITAL PRODUCTION and was written by Bela Beier.

A recent study from the Beijing Institute of Technology and Tsinghua University Institute for Artificial Intelligence introduces ActiveSplat, an autonomous high-fidelity reconstruction system. This innovative system, detailed in a research paper , uses Gaussian Splatting to build 3D maps in real-time, providing high-accuracy results for VFX, game development, and digital production.

ActiveSplat: A Hybrid Approach to 3D Mapping

ActiveSplat leverages a hybrid mapping strategy that combines dense environmental data with sparse abstraction of the space. Using a Gaussian map and Voronoi graph, the system enables “efficient and complete exploration” with high-fidelity reconstruction. This combination of sparse topology for viewpoint sampling and path planning, alongside dense prediction for view-dependent selection, allows the system to make efficient decisions with promising accuracy.

The research highlights how ActiveSplat’s hierarchical planning strategy, based on a topological map, minimizes repetitive trajectories during 3D scanning. Additionally, the system can adjust its yaw and pitch to handle areas that are difficult to reach or where data might be lost, improving its adaptability to different environments. These advancements position ActiveSplat as a potentially powerful tool for future virtual production workflows, offering a more efficient way to integrate real-world environments into digital projects.

Research and Development Outlook

While this research represents an exciting leap forward in 3D spatial mapping, the system is still in the development phase. ActiveSplat’s potential to simplify VFX and game development processes is significant, but it requires further refinement before widespread implementation in production. For now, the study serves as a valuable exploration of how AI and spatial computing could redefine mapping technologies in digital production, with more updates to come. To stay informed, visit the full (and free) article of the researchers – well worth a read!

p.s.: What is Gaussian Splatting?
Gaussian splatting is a technique used to capture and reconstruct real-world objects in 3D by blending rasterization and optimization. Unlike traditional triangle-based meshes, this method uses volume-based representation, with 3D data defined by Gaussian splats that include position, covariance, color, and alpha channels. As a newer technology, Gaussian splatting enables real-time photo realism and direct editing, with applications in VR, AR, and emerging production pipelines.

The post Study Explores Automated 3D Mapping for VFX and Game Development first appeared on DIGITAL PRODUCTION and was written by Bela Beier.