A Billion 3D Splats Rendering in Your Browser (And It's Open Source)

Summary

The video highlights recent improvements to 3D Gaussian Splatting (3DGS), especially the ability to display extremely large scenes—reportedly up to one billion splats—in a web browser. The presenter says scenes that once overwhelmed consumer computers can now run on laptops and phones, without requiring users to download a dedicated viewer.

Technology and open-source tools

• Splats are small scene elements that encode properties such as position, shape, and color. They can produce highly realistic captures, but large scenes have traditionally required substantial storage and computing power.
• AholoViewer is an open-source browser viewer released under the MIT license. The presenter describes it as suitable for embedding in websites or adapting for custom applications.
• Supporting command-line tools can prepare and optimize splat scenes. The presenter highlights support for SPZ, a compact format, as well as conversion and optimization tools.
• The toolkit can create level-of-detail (LOD) chunks and stream them as needed. LOD reduces detail for distant areas, while streaming progressively loads portions of a scene instead of rendering the entire large file at once.
• It can also generate collision geometry, allowing users to move through scanned spaces in a game-like way.

Demonstrations and practical guidance

The presenter runs the viewer locally and shows several examples, including drone-captured scenes, a berry scan, a castle, and an apartment interior. They report that the examples load quickly and that their local tests ran at around 120 FPS. A berry scan shared by Danny Bittle is noted as being available under a CC4 license.

The video also demonstrates the general workflow: capture a location or object using, for example, a phone app, video, or drone footage; process and optimize the resulting splats; create streamed LOD chunks and collision geometry if needed; then load the result in a viewer or game project. The video is a practical overview and demonstration rather than a detailed, step-by-step installation tutorial.

Use cases and assessment

The presenter identifies real-estate interiors as an immediate fit and sees growing potential for game assets and walkable scanned environments. Other suggested uses include historical sites, landmarks, and interactive scans embedded on websites. They also mention that Unreal Engine is pursuing a related approach to rendering splats, comparing it to a Nanite-like system.

Overall, the video argues that advances in streaming, LOD, file optimization, and collision generation are making 3DGS more practical—not just to view, but to use in interactive applications.

Main speakers and sources

• Stefan provides the commentary, tool overview, and local demonstrations.
• Aholo / AholoViewer is the open-source viewer and associated splat-processing toolset discussed.
• Danny Bittle is cited as the creator of the berry scan example.
• Unreal Engine is mentioned as an example of a related rendering direction.