Gaussian Splatting Software in 2026: How to Choose

What gaussian splatting software does

Gaussian splatting software turns a set of overlapping photos, or a video, into a 3D scene you can orbit and fly through. The scene is made of a very large number of small, soft, colored blobs called Gaussians. The method comes from the paper 3D Gaussian Splatting for Real-Time Radiance Field Rendering by Kerbl, Kopanas, Leimkühler and Drettakis, published at SIGGRAPH 2023.

People search for the best gaussian splatting software, but there is no single answer. The tools differ in where they run, who they are built for and how you pay. This overview groups them into four types: phone apps, desktop apps, open-source trainers, and editors and viewers.

Every tool does roughly the same two jobs. First it works out where each photo was taken from, a step called structure-from-motion. Then it trains the Gaussians until they reproduce your photos. Some tools do both steps for you, and some expect you to bring the camera positions yourself.

Everything below was checked on each vendor's own site or code repository in October 2026. Plans and features change, so confirm on the vendor page before you pay for anything. Third-party prices are left out for that reason; only the pricing model is described.

Phone apps: capture and process in one place

A gaussian splatting app on a phone is the shortest route from nothing to a first scene, because the camera and the software are the same device.

Scaniverse, from Niantic Spatial, runs on iOS and Android. Its product page says Gaussian splats and meshes are processed on-device. The pricing page lists a free plan and says you can continue using Classic Scaniverse to generate unlimited 3D Gaussian splats on your device. The newer Scaniverse plans work on monthly credits, with paid subscription tiers, and some of their assets are uploaded for processing. Listed exports include PLY and Niantic's own SPZ format. It suits people who want a free start with nothing more than a phone.

Polycam lists iOS, Android and web browsers as platforms. Its pricing page shows a free plan with limited captures and paid subscription plans, and states plainly that there is no one-time purchase option. Its Gaussian splatting page accepts uploads of 20 to 200 photos, or a video. The pages read for this article do not say where processing happens. It suits people who want one app for several kinds of 3D scanning.

KIRI Engine is available for iPhone, iPad, Android and the web. Its pricing page describes a free Basic plan for photo and LiDAR scanning and a Pro subscription; the 3D Gaussian splatting scan, which the site describes as including a mesh, is listed under Pro. The same page refers to servers that process your scans. It suits people who want a mesh alongside the splat.

Desktop apps: train on your own computer

Desktop apps use your computer's graphics card instead of a phone or a remote server. Your files stay on your machine, and you can use photos from any camera.

Jawset Postshot is a Windows application. Its site lists the system requirements as Windows 10 or later and an NVIDIA GPU, naming the GeForce RTX 2060 or higher, and says all processing and machine-learning steps run on devices only you have access to. It accepts photos and video, and the site lists Unreal Engine and After Effects integrations. The pricing page shows a free plan limited to non-commercial use, and paid Indie and Studio plans with monthly or yearly billing. It suits Windows users with an NVIDIA card, especially those working in video or game engines.

SplatLoom is a desktop app for Mac (Apple Silicon, macOS 15 or later) and Windows 10/11 x64. You pick photos or a video and press Generate; camera solving and training are bundled, so no Python or command line is needed. It runs entirely on your computer with no upload and no account. It is a one-time purchase of US$19.90 that includes both installers. It exports Gaussian PLY, a PNG still, an MP4 orbit video and an offline HTML viewer. It suits Mac users and beginners who want a short path, and its limits are real: it has fewer controls than Postshot or the open-source trainers, its Windows GPU support is newer than its Mac support, and it does not export meshes.

Free gaussian splatting software: open-source trainers

If you are comfortable with a terminal, the open-source trainers cost nothing and expose every setting.

The Inria reference implementation is the original code from the paper. Its README asks for a CUDA-ready GPU and mentions 24 GB of VRAM to train to the paper's evaluation quality, and says it was tested on Windows 10 and Ubuntu Linux 22.04. One detail is easy to miss: its licence allows research and evaluation use only, and commercial use is not permitted. It suits researchers and people who want the reference behavior.

gsplat is an open-source library for CUDA-accelerated rasterization of Gaussians with Python bindings, under the Apache-2.0 licence. nerfstudio, also Apache-2.0, uses gsplat as the backend for its own Gaussian splatting method, called Splatfacto, which you run with one command and can export as a PLY. The nerfstudio README says you must have an NVIDIA video card with CUDA installed. These suit people who already work in Python.

Brush is an Apache-2.0 trainer built on WebGPU-compatible technology. Its README says it works on macOS, Windows and Linux, with AMD, NVIDIA and Intel cards, on Android and in a browser, and the releases page offers prebuilt downloads for Apple Silicon Macs, 64-bit Windows and 64-bit Linux. It takes COLMAP data or Nerfstudio-format datasets as input, so you prepare camera positions first, typically with COLMAP. It suits people without an NVIDIA card who do not mind that extra step.

Editors and viewers

Training is rarely the last step. Most scenes have stray blobs around the edges that need removing, and the result has to be shared somehow.

SuperSplat, from PlayCanvas, is the main free tool for this. It is an MIT-licensed, browser-based editor: you load a splat file, select and delete unwanted Gaussians, crop, move, rotate and scale, then export or publish. Its documentation says everything runs in the browser and nothing is uploaded until you choose to publish. It needs a browser with WebGPU. It exports PLY, compressed PLY, SOG, SPZ and a standalone HTML viewer.

SuperSplat does not create splats. PlayCanvas says in its documentation that it does not provide a tool for converting photos into a splat, so it pairs with any of the tools above. Most of them can produce a Gaussian PLY, which is the format SuperSplat and most other viewers accept.

How to choose by device

Your hardware removes most options before taste comes into it.

• Phone only: start with a phone app. Scaniverse processes on the device; Polycam and KIRI Engine also offer web versions.
• Mac with Apple Silicon: SplatLoom, or Brush if you are willing to prepare camera positions yourself. Postshot is Windows-only, and the Inria and nerfstudio trainers expect an NVIDIA GPU with CUDA.
• Windows PC with an NVIDIA RTX card: this is where you have the widest choice. Postshot, SplatLoom, Brush, nerfstudio and the Inria code are all possible.
• Windows PC with an AMD or Intel card: Brush lists support for these cards. SplatLoom uses Brush for training, but its Windows GPU support is newer than its Mac support, so check that it runs on your machine; the refund policy covers the case where it does not.
• Linux: Brush, nerfstudio and the Inria code. SplatLoom and Postshot do not offer Linux versions.

How to choose by goal

What you want to do with the result matters as much as the device.

• A quick first scene to see what the technique looks like: a phone app.
• Keeping files on your own device: a tool that processes locally. Scaniverse's on-device mode on the phone; Postshot, SplatLoom or the open-source trainers on a computer.
• Photos from a real camera or a drone: a desktop app or an open-source trainer, where you choose the input files.
• Full control over training settings, or research: gsplat, nerfstudio or the Inria code.
• Commercial work: read the licence first. The Inria code is for non-commercial research and evaluation, and Postshot's free plan is for non-commercial use. gsplat, nerfstudio and Brush are Apache-2.0.
• A mesh for 3D printing or CAD: a Gaussian splat is not a mesh, and SplatLoom does not export one. Look at a tool that produces meshes, such as KIRI Engine, Scaniverse or Polycam, all of which list mesh output on their sites.
• Cleaning up and publishing on the web: SuperSplat, whatever you trained with.

Whichever tool you pick, capture matters more than software. Reflective, transparent and moving subjects are hard for all of them, and gaps in your coverage show up as holes or haze.

Where SplatLoom fits

SplatLoom sits in one corner of this map: a local desktop app for people who do not want to install Python, set up COLMAP or create an account. You choose photos or a video, press Generate, tidy the scene with box or sphere crop and one-click spike removal, and export.

As a rough guide to speed, in single test runs on an Apple M5 Pro MacBook at default settings, phone videos of 13 to 20 seconds took about 9 to 14 minutes from import to a finished scene. Typical photo sets are 30 to 150 overlapping photos. Other hardware will differ.

It is not the right choice if you need meshes, measurements, Linux, or fine control over training. For those, the tools above are a better fit. And if you already have a splat from any of them, the free viewer at splatloom.com/ply-viewer opens Gaussian .ply and .sog files locally in the browser.