A place to keep
the light
rvrry is named for reverie — that soft, drifting state between memory and dream. It's where we think captured moments belong.
A Gaussian splat is a strange kind of photograph. Instead of a flat grid of pixels, it's a cloud of millions of soft, coloured points floating in space — a scene reconstructed as light itself. You don't look at it. You move through it.
The trouble is that these dreams are awkward to keep. The files are large, the formats are unfamiliar, and most viewers feel like engineering tools — sharp edges, loading bars, jargon. The beauty gets lost on the way to the screen.
rvrry exists to fix that one quiet thing — and to widen the path on both ends. Hand it a video or a set of images and it reconstructs them into a splat. Then it compresses and optimizes the scene so it streams fast, gives it a home, wraps it in a viewer that's soft and calm, and hands you a single link. No app to install. No account to wrestle. Just the scene, drifting gently, the way you remember it.
Because splats can hold whole spaces, not just objects, rvrry is built for world‑scale 3D — rooms, buildings, sites — and for digital twins: living, measurable replicas of real places you can revisit and watch change over time.
For the people mapping their world in light
Artists, photogrammetrists, researchers, archivists, surveyors, and anyone who pointed a phone at something they didn't want to forget. If you can capture it, rvrry can reconstruct, shrink, and hold it — and make it feel like a place worth returning to.
Built with
rvrry's viewer renders Gaussian splats in real time using WebGL via the open‑source gsplat.js library, so scenes play directly in the browser with no plugins. Compression and optimization trim, quantize, and reorder the point data for fast progressive streaming and level‑of‑detail. The site itself is plain, fast HTML, CSS, and JavaScript — no framework, no build step — because dreams shouldn't need a toolchain to load.
It works with video and image inputs as well as the common splat formats — .splat, .ply, and compressed .ksplat — produced by pipelines like the original 3D Gaussian Splatting trainer, nerfstudio, Luma, and Polycam.
The main set of these features is rolling out soon. If you'd like to be there when the dream opens, leave your details.