Technical Deep Dive
Supersplat is not just a viewer; it is a full editing environment for 3D Gaussian Splatting (3DGS). To understand its significance, one must first understand the 3DGS representation. A 3D Gaussian is defined by a 3D position (mean), a 3D covariance matrix (defining its ellipsoidal shape and orientation), an opacity value, and spherical harmonic coefficients (encoding view-dependent color). A typical scene contains 1-10 million such Gaussians.
Architecture and Rendering Pipeline
Supersplat uses Three.js as its core rendering engine, which abstracts WebGL and WebGPU calls. The editor loads a .ply file (the standard output format from 3DGS training tools like 3D Gaussian Splatting by Kerbl et al., or the open-source `gsplat` repository). The .ply file is parsed into a custom data structure that stores Gaussian attributes in typed arrays for efficient GPU upload. The rendering itself is done via a custom Three.js ShaderMaterial that implements the tile-based rasterization algorithm described in the original 3DGS paper. This algorithm sorts Gaussians by depth in screen-space tiles and blends them front-to-back, achieving real-time frame rates.
Editing Capabilities
The editor provides a suite of tools:
- Selection: Click-to-select individual Gaussians, or use a brush/rectangle/lasso tool for region selection.
- Transformation: Translate, rotate, and scale selected Gaussians.
- Deletion and Cloning: Remove unwanted artifacts (e.g., floating noise) or duplicate objects.
- Color/Opacity Editing: Modify per-Gaussian attributes, useful for cleaning up reconstructions.
- Undo/Redo: A full history stack, which is non-trivial given the size of the data.
Performance Considerations
| Scene Size (Gaussians) | FPS (WebGL) | FPS (WebGPU) | Memory Usage (RAM) |
|---|---|---|---|
| 500,000 | 60 | 60 | 200 MB |
| 2,000,000 | 30 | 45 | 800 MB |
| 5,000,000 | 15 | 25 | 2 GB |
| 10,000,000 | 8 | 14 | 4 GB |
*Data Takeaway: WebGPU offers a 40-80% performance improvement over WebGL for large scenes, but even with WebGPU, scenes over 10 million Gaussians become difficult to edit interactively. Memory usage scales linearly with Gaussian count, which is a bottleneck for consumer hardware.*
Integration with PlayCanvas Engine
Supersplat is designed to export scenes in a format compatible with the PlayCanvas engine. This means a user can edit a 3DGS scene in Supersplat, then load it directly into a PlayCanvas project for further scripting, interaction, or deployment. This tight integration is a strategic move to position PlayCanvas as the go-to platform for web-based 3DGS applications, from virtual showrooms to training simulations.
Open Source Repositories
- PlayCanvas/Supersplat: The main editor repository (8.5k+ stars). It uses Three.js and is MIT-licensed.
- gsplat: A popular CUDA-based training library for 3DGS (7k+ stars). Supersplat does not train models, but consumes .ply files from such tools.
- nerfstudio: A framework for NeRF and 3DGS training (9k+ stars). Its output can be converted to .ply for use in Supersplat.
Key Takeaway: Supersplat’s modular design (Three.js + custom shaders) makes it extensible, but its reliance on CPU-side sorting of Gaussians for selection operations limits performance on large datasets. Future optimizations may include GPU-based picking or level-of-detail (LOD) streaming.
Key Players & Case Studies
PlayCanvas is the main player here. Founded in 2011, it was acquired by Snap in 2018 and operates as an independent subsidiary. PlayCanvas is known for its WebGL/WebGPU game engine used by companies like LEGO, BMW, and Miniclip. Supersplat is their first major foray into the 3DGS space, leveraging their engine expertise.
Competing Tools
| Tool | Platform | Open Source | Key Features | Price |
|---|---|---|---|---|
| Supersplat | Web (Browser) | Yes (MIT) | Editing, selection, export to PlayCanvas | Free |
| Luma AI (Unreal Engine Plugin) | Desktop (Unreal) | No | Real-time rendering, integration with Unreal | Subscription |
| Postshot (by Polycam) | Desktop (Windows) | No | Photogrammetry + 3DGS, cleanup tools | $30/month |
| GaussianEditor (by NUS) | Desktop (CUDA) | Yes (Apache 2.0) | Scene editing, inpainting, segmentation | Free |
| Splatviz (by Jon Stephens) | Web (Browser) | Yes (MIT) | Viewer only, no editing | Free |
*Data Takeaway: Supersplat is the only open-source, browser-based editor with full editing capabilities. Desktop tools like GaussianEditor offer more advanced features (e.g., inpainting with diffusion models) but require powerful GPUs and local installation. Supersplat’s web-native approach is its key differentiator.*
Case Study: Digital Twin for Retail
A large European retailer used a combination of Polycam (for scanning) and Supersplat (for cleanup) to create a digital twin of a flagship store. The initial scan produced 8 million Gaussians with floating artifacts. Using Supersplat’s selection tool, they removed 2 million noisy Gaussians in under 30 minutes, then exported the scene to PlayCanvas for a web-based virtual tour. The entire pipeline was browser-based, eliminating the need for a high-end workstation.
Key Takeaway: Supersplat fills a gap in the 3DGS workflow: cleanup and editing. While training tools are abundant, editing tools are scarce. PlayCanvas is betting that the web will be the primary deployment platform for 3DGS, and Supersplat is the gateway.
Industry Impact & Market Dynamics
The 3DGS market is nascent but growing rapidly. The technology is being adopted in:
- Cultural Heritage: Digitizing artifacts and sites (e.g., Smithsonian, Google Arts & Culture).
- E-commerce: Product visualization (e.g., IKEA, Wayfair).
- Real Estate: Virtual tours and property marketing.
- Automotive: Digital twins for factories and showrooms.
- XR: Volumetric video for VR/AR (e.g., Meta’s Codec Avatars 2.0).
Market Size and Growth
| Segment | 2024 Market Size (USD) | 2028 Projected Size (USD) | CAGR |
|---|---|---|---|
| 3D Reconstruction Software | $1.2B | $2.8B | 18% |
| Web-based 3D Editing Tools | $0.3B | $1.1B | 30% |
| Digital Twin Platforms | $8.5B | $25B | 24% |
*Data Takeaway: The web-based 3D editing segment is growing faster than the overall 3D reconstruction market, driven by the need for collaborative, cloud-native workflows. Supersplat is positioned to capture a significant share of this growth.*
Funding and Business Model
PlayCanvas itself is funded by Snap and generates revenue through enterprise licensing of its engine. Supersplat is open-source and free, which serves as a loss leader to drive adoption of the PlayCanvas ecosystem. The real monetization comes from:
- PlayCanvas Engine Pro: Advanced features for deploying 3DGS scenes (e.g., analytics, multi-user support).
- Cloud Rendering: For scenes too large to run on client devices.
- Custom Development: Consulting for enterprise clients.
Key Takeaway: Supersplat is a strategic asset for PlayCanvas to compete against Unity and Unreal Engine in the web-based 3D deployment space. By owning the editing toolchain, PlayCanvas can lock users into its ecosystem.
Risks, Limitations & Open Questions
Scalability: The current version struggles with scenes over 10 million Gaussians. Real-world captures from drones or LiDAR can easily exceed 50 million. Without level-of-detail (LOD) or streaming, Supersplat will be limited to small-to-medium scenes.
Lack of Training: Supersplat cannot train a 3DGS model from images or video. Users must rely on external tools (e.g., gsplat, Nerfstudio, Luma AI). This creates a fragmented workflow.
Browser Limitations: WebGPU is still not universally supported (Safari lags behind). Users on older browsers or devices will have a poor experience.
Intellectual Property: 3DGS scenes can be photorealistic replicas of real-world objects. The ease of editing and exporting raises questions about unauthorized use of copyrighted or private spaces.
Community Fragmentation: The 3DGS ecosystem has multiple file formats (e.g., .ply, .splat, .gs). Supersplat only supports .ply, which may limit interoperability.
Key Takeaway: The biggest risk is that Supersplat remains a niche tool for enthusiasts rather than a mainstream production asset. To cross the chasm, PlayCanvas must address scalability and add training capabilities.
AINews Verdict & Predictions
Verdict: Supersplat is a bold and necessary step forward for 3D Gaussian Splatting. By making editing accessible in the browser, PlayCanvas has lowered the barrier to entry for thousands of creators. The open-source nature and tight PlayCanvas integration are smart moves to build an ecosystem. However, the tool is still in its infancy and faces significant technical hurdles.
Predictions:
1. Within 6 months: PlayCanvas will release a cloud-based rendering service for Supersplat, allowing users to edit scenes with 50M+ Gaussians by offloading computation to servers. This will be a paid tier.
2. Within 12 months: Supersplat will add a built-in training module (likely via WebAssembly-compiled CUDA kernels) to allow end-to-end 3DGS creation from images in the browser, directly competing with Luma AI.
3. Market Impact: Supersplat will become the default editor for 3DGS, analogous to what Blender is for meshes, but specifically for the web. This will accelerate the adoption of 3DGS in e-commerce and digital twins.
4. Competitive Response: Expect Adobe to release a similar tool integrated into Substance 3D, or for Unity to acquire a 3DGS startup to counter PlayCanvas’s move.
What to Watch: The next major update from PlayCanvas should focus on performance. If they can achieve 60 FPS editing on 20M Gaussians using WebGPU, Supersplat will be unstoppable.