Press & Media

3D Gaussian Splatting, native on the Mac. No cloud, no NVIDIA, no command line.

RadianceKit turns a handful of photos or a short video into a photorealistic 3D scene — computed entirely on your own Mac, on the Apple Silicon GPU. This page collects everything press and reviewers need. For a review code or the full asset bundle, just email info@kindler-dev.de.

Downloads

At a glance

ProductRadianceKit
What it isNative macOS app for 3D Gaussian Splatting — train, view and export on-device
CategoryGraphics & Design (Mac App Store)
PlatformmacOS 26 (Tahoe) or later, Apple Silicon (M1 or later)
Current version1.8.0 (July 2026)
PriceFree 3-day trial, then a one-time in-app purchase — US$7.99 / 8.99 € — no subscription
LanguagesInterface and manual in 18 languages
Export6 file formats (PLY, compressed PLY, SOG, SPZ, glTF/GLB, .splat) plus orbit video, wigglegram and web viewer · publishing to SuperSplat
Privacy100% local — images never leave the Mac
Mac App StoreDownload
DeveloperBjörn Kindler (independent developer, near Munich, Germany)

About RadianceKit

Gaussian Splatting is one of the most exciting 3D reconstruction techniques of recent years — yet until now it has been reachable almost exclusively through NVIDIA GPUs, the cloud, or cryptic command-line tools. RadianceKit changes that: it trains complete Gaussian Splatting scenes directly on the Mac, via Apple's Metal GPU compute, inside a clean native interface. Users drop in a folder of images or a video, the integrated Structure-from-Motion pipeline recovers the camera poses, and training runs with a live preview in the viewport. Two scientifically grounded training strategies (Classic and MCMC), validated quality presets, and WYSIWYG export to six formats make the result immediately usable — from web viewer to 3D pipeline. Everything stays on the device.

Why it's different

  • Fully native on Apple Silicon — Metal GPU compute, no CUDA, no Python, no Docker.
  • On-device — images stay on the Mac; no uploads, no compute subscription.
  • Two strategies — Classic (Kerbl et al. 2023) and MCMC (Kheradmand et al. 2024, NeurIPS Spotlight).
  • Validated presets — quality tiers and Capture-Class presets (Drone/Aerial, 360° Walkaround, Photo/Object) tuned across 560+ documented experiments.
  • Real-time viewport — live preview, plus an in-house renderer with true WYSIWYG between preview and export.
  • Integrated pipeline — built-in SfM and six export formats.
  • Native 360° ingest — equirectangular, stereo 360°, cubemap, equi-angular cubemap, fisheye 180/200° and partial panoramas, detected from metadata and unwrapped into perspective views automatically.
  • The complete chain in one app — from the source images through camera solving to the finished splat. Other tools also train locally on a Mac, but expect camera poses solved in a second program.
  • Scene editor — clean up a finished scene without leaving the app: brush selection, cylinder crop, editing slab, colour matching, multi-step undo.

New in version 1.8.0

  • Scene editor: brush-select regions and delete them, optionally with sharp edges that cut cleanly through partially covered splats. Cylinder crop for drone and walkaround captures, a movable editing slab, colour matching against a reference area you paint yourself, multi-step undo.
  • Markedly more accurate camera positions: the app's own solver made its largest accuracy gain since launch and now aligns entire capture sequences that previously defeated it. The new "Maximum Quality (Native)" preset bundles the most thorough configuration.
  • Publishing finished scenes to SuperSplat, reading back the app's own compressed PLY files.
  • Detection of printed AprilTag markers in the footage.
  • Inspector reordered to follow the actual workflow; unavailable settings stay visible.
  • Steadier viewport.

Context: where Gaussian Splatting stands today

The technique has moved from research into practice. Framestore used 4D Gaussian Splatting for the hologram sequences in James Gunn's Superman (2025); in heritage documentation it is used to record the state of a site quickly and in a form that runs in a browser — peer-reviewed studies show phone capture is sufficient and that delivery is roughly forty times lighter than laser scanning. Where millimetres matter, however, Gaussian Splatting replaces neither photogrammetry nor laser scanning. It is a medium for viewing and communicating, not for survey.

About the developer

RadianceKit is the work of Björn Kindler, an independent software developer based near Munich, Germany. The app grew out of more than 560 documented training experiments, with the goal of making a research-born technique genuinely usable for Mac users — without specialist hardware and without the command line. RadianceKit is self-funded.

"3D graphics have fascinated me ever since I tinkered with GigaCAD on the C64 as a teenager, then worked my way from Quicksilver to TurboSilver and Imagine on the Amiga. That thread runs all the way to today. Gaussian Splatting finally makes photorealistic 3D from ordinary photos possible — but every path to it ran through NVIDIA cards, cloud services or the command line. I wanted my Mac to just do this: drag in some photos, get a photorealistic 3D model back, and have nothing leave the machine. That's exactly what RadianceKit is." — Björn Kindler

Press contact

Kindler Dev — Björn Kindler · info@kindler-dev.de · +49 89 62084444

We're happy to provide review codes with full functionality to editors.

Apple, Apple Silicon, Mac, macOS and Metal are trademarks of Apple Inc., referenced descriptively; no affiliation with or endorsement by Apple is implied.