Scantic trains 3D Gaussian Splats entirely on your phone
Software developer Sebastian Beyer has released Scantic, a new iPhone app for generating 3D Gaussian Splats of objects and environments from standard phone camera footage.
Training takes “under a minute”, and unusually for apps of this type, is performed locally on the phone – the scan data isn’t uploaded to the cloud for processing.
The base app is free, although you can pay to unlock a higher-quality export mode.
Generate 3D Gaussian Splats of real objects and environments from phone footage
Scantic part of a growing set of 3D Gaussian Splatting tools: a new way to reconstruct real-world objects from source images or video, creating high-quality, fast-rendering 3D representations.
As with other such tools, users simply start a new scan, then move in circles around the object or environment being scanned, before Scantic generates the splat from the images captured.
Processing takes “under a minute [for] smaller scans”, and the Splat can be exported in PLY format, for use in DCC applications that support 3DGS data.
You can judge the quality of the output from the video at the top of the story.
Runs entirely locally on your iPhone, without any cloud processing
We’ve covered other mobile 3D scanning apps that can generate 3D Gausssian Splats on CG Channel in the past, including Kiri Engine and Polycam.
Both are more fully featured, and come with tools for cropping and manipulating scan data.
However, the unusual thing about Scantic is that processing is done entirely locally on the user’s phone: the source data isn’t uploaded to the cloud for processing.
As well as avoiding any potential security issues, that cuts out a time-consuming step: Beyer says that Scantic can train a Splat in the time it takes other apps just to upload the data.
Created as a personal development project
Beyer’s background is in software engineering rather than CG art, although he uses Blender for personal work.
Scantic came out of a self-imposed summer project: to improve the performance of an existing 3D Gaussian Splatting workflow using AI coding tools, without being a 3DGS expert himself.
His first Splat, created using the open-source Brush engine from footage captured on his iPhone, took over three hours to train on a MacBook Air.
However, with the help of the Claude Fable 5 AI model, he steadily improved performance – eventually, by up to 260x – then began to move larger and larger parts of the pipeline to the iPhone itself, eventually ending up with an iPhone-only app “almost by accident”.
May get a macOS companion app, but no plans for an Android edition
In future, Beyer aims to make it possible to use Scantic to train Splats from existing videos rather than footage captured live, and to export Splats in SPZ format.
However, while he may release a macOS companion app, to transfer a scan to a laptop or desktop system for processing, he currently has no plans for an Android version.
“[With Apple products] the hardware range is relatively controlled,” he says. “With Android [the app would probably only run] on certain high-end devices while the rest would get a bad user experience.”
Price and system requirements
Scantic is compatible with iOS 17.0+ and requires a device with an A12 Bionic chip or better, so it should run on a 2018 iPhone XS or newer.
The base app is free, and provides access to the standard quality export mode, which generates a thinned-down, lower-quality version of the scan.
Full quality export requires a paid subscription, which costs €29.99/year (around $35/year). You can also pay to export individual scans at full quality, which costs €4.99 per scan.
Read more about 3D Gaussian Splatting app Scantic on the product website
Download Scantic from the App Store
Have your say on this story by following CG Channel on Facebook, Instagram and X (formerly Twitter). As well as being able to comment on stories, followers of our social media accounts can see videos we don’t post on the site itself, including making-ofs for the latest VFX movies, animations, games cinematics and motion graphics projects.