SuperSplat: Turn Phone Photos Into Photorealistic 3D Walkthroughs (Open Source)

By Prahlad Menon 5 min read

There’s a tweet going viral right now with 1.3 million views and a simple claim: AI reconstructs reality from your photos, the result loads on a phone, and it looks like you’re actually there.

The technology is Gaussian Splatting. The tool is SuperSplat, built by PlayCanvas. And the business angle is what’s making people pay attention — freelancers are already charging $300–$800 per scan for realtors, Airbnb hosts, venues, car dealers, and museums.

One person. One phone. One weekend. A business.

What Is Gaussian Splatting?

Traditional 3D reconstruction (photogrammetry, NeRFs) has been around for years, but it was slow, required expensive hardware, and produced results that looked obviously synthetic. Gaussian Splatting changed the game in 2023 by representing 3D scenes as collections of 3D Gaussians instead of meshes or neural fields.

The result: photorealistic reconstructions that render in real-time on consumer hardware — including phones. You take photos of a space from multiple angles, feed them through a reconstruction pipeline, and get an interactive 3D environment you can walk through in your browser.

It looks like you’re there because, geometrically, you almost are. The reconstruction captures lighting, reflections, and spatial relationships with a fidelity that previous approaches couldn’t match at real-time framerates.

SuperSplat: The Open-Source Editor

SuperSplat is PlayCanvas’s free, open-source tool for inspecting, editing, optimizing, and publishing 3D Gaussian Splats. It runs entirely in the browser — nothing to download or install.

What it does:

  • Load and inspect Gaussian Splat scenes
  • Edit and clean up captures (remove artifacts, crop regions)
  • Optimize file sizes for web deployment
  • Publish directly to shareable URLs
  • Browse and download community-created splats

The tech stack:

  • Built on PlayCanvas, the open-source WebGL game engine
  • Runs in any modern browser
  • No server-side processing for viewing
  • MIT License (GitHub repo)

The editor itself doesn’t do the initial reconstruction from photos — you need a separate tool for that (like Nerfstudio, COLMAP, or commercial options like Luma AI or Polycam). SuperSplat handles everything after capture: editing, optimization, and publishing.

The Freelance Opportunity

This is what’s driving the viral attention. The workflow for turning this into a service business:

  1. Capture — Walk through a space with your phone, taking overlapping photos (or video)
  2. Reconstruct — Run the photos through a Gaussian Splatting pipeline
  3. Edit — Clean up artifacts in SuperSplat, crop to the relevant space
  4. Publish — Deploy as an embeddable web experience

The output is an interactive 3D walkthrough that loads on any device with a browser. No app download required. No VR headset. Just a link.

Who’s paying for this:

  • Realtors — Virtual open houses that feel like being in the property
  • Airbnb hosts — Immersive previews that increase booking confidence
  • Event venues — Walk the space before committing to a booking
  • Car dealerships — Interior/exterior walkthroughs without visiting the lot
  • Museums and galleries — Virtual exhibitions accessible worldwide
  • Construction/architecture — Progress documentation and client presentations

At $300–$800 per scan, a freelancer doing 2-3 scans per week is generating $30K–$100K annually from a phone and free software.

Why This Is Different From Virtual Tours

Google Street View-style tours use stitched 360° photos. You jump between fixed positions and look around. It works, but it feels like looking at photos — because you are.

Gaussian Splatting creates a continuous 3D space. You can move smoothly through it, look at objects from angles the photographer never explicitly captured, and perceive depth naturally. The difference is immediately obvious when you see both side by side.

Matterport has been doing premium 3D tours for years, but requires a $3,000+ camera and charges monthly software fees. The Gaussian Splatting workflow produces comparable (often superior) visual quality from phone photos and free software.

The Full Pipeline (Open Source)

Here’s how to go from photos to published 3D walkthrough using entirely free tools:

  1. Capture: Phone camera (50-200 photos of the space, overlapping coverage)
  2. Structure from Motion: COLMAP (determines camera positions)
  3. Gaussian Splatting: gsplat or 3D Gaussian Splatting (creates the 3D representation)
  4. Editing: SuperSplat (clean, crop, optimize)
  5. Publishing: SuperSplat’s publish feature or self-host with PlayCanvas viewer

Total cost: $0 in software. A modern GPU helps for step 3 (reconstruction), but cloud GPU instances work fine for occasional use.

Market Context

The 3D scanning market is projected to exceed $10 billion by 2028. Most of that has been dominated by expensive hardware (Matterport, Leica, FARO) and proprietary software. Gaussian Splatting democratizes the capture side, and SuperSplat democratizes the editing/publishing side.

What remains is the service layer — someone who knows how to capture well, clean up the results, and deliver a polished product. That’s the freelance opportunity. The technology is free; the expertise and execution are what clients pay for.

Getting Started

To try SuperSplat immediately:

To set up local development:

git clone https://github.com/playcanvas/supersplat.git
cd supersplat
npm install
npm run develop
# Navigate to http://localhost:3000

To learn the full pipeline:

The Bigger Picture

Gaussian Splatting is doing to 3D what the iPhone did to photography: making professional-quality output accessible to anyone with consumer hardware. The quality gap between a $50,000 Matterport setup and a phone-based Gaussian Splat workflow is narrowing every month.

SuperSplat being 100% open source means the tools won’t get paywalled (PlayCanvas has been MIT-licensed since 2014). The only gatekeeper left is skill — and skill is learnable over a weekend.

Links: