AirVis Studio Creates Gaussian Splats Locally for Free

How it works

A creator can start with photos, standard video, 360 video, or COLMAP data, which already contains estimated camera positions and source images.

Studio estimates camera positions when needed, then uses the computer’s graphics processor to train a Gaussian splat. Millions of small, colored points make the scene viewable from new positions.

Studio can mask people or isolate an object before training. It can also make scenes load in stages and give floors, walls, and stairs solid boundaries.

A finished scene can stay on disk or be published through AirVis (opens in a new tab). The wider AirVis app can open shared scenes across phones, computers, Apple Vision Pro, Meta Quest, and Android XR.

Hardware

The Mac version requires macOS 26 or later, Apple silicon, 16 GB of unified memory, and 100 GB of storage for project data.

On Windows, Studio supports 64-bit Windows 10 and 11 with a Vulkan 1.1 graphics processor. AirVis recommends 8 GB of video memory, 32 GB of system memory, and 100 GB of storage. The company says the app does not require a graphics card from one specific vendor.

Larger scenes still need more memory, processing time, and graphics power. AirVis has not published training times or performance comparisons across Intel, AMD, NVIDIA, and Apple hardware.

Availability

AirVis Studio has no capture credits or scene limit, and storage is limited only by the creator’s disk. It is separate from the broader AirVis app, where cloud processing, sharing, and a paid Pro plan remain optional.