Upload 1–10 photos. Pick one and a depth-estimation model running locally in your browser turns it into a cloud of colored 3D splats you can orbit, zoom and fly through. Nothing is uploaded anywhere.
Your browser doesn't support WebGL2, which this splat renderer needs. Try a recent version of Chrome, Firefox or Edge.
Gaussian splatting is a way of representing a 3D scene not as triangles but as a cloud of soft, colored blobs — gaussians — each with its own position, size and color, blended together from whichever direction the camera looks. Introduced in 2023, it became popular because scenes built this way render at very high frame rates while still capturing the soft edges, fine detail and reflections that triangle meshes struggle with, and because a scene can be produced directly from ordinary photographs rather than modeled by hand.
Building a full gaussian-splat scene the way researchers do normally starts with dozens or hundreds of photos taken from every angle around a subject, a separate step to work out exactly where each photo was taken from, and several minutes of GPU optimization to fit the gaussians to those views. This page takes a lighter, single-photo route instead: a depth-estimation model looks at one image and estimates how far away every part of it is, and that estimate is enough to lift the picture off the flat plane into a real, orbitable 3D cloud of splats — closer to the "2.5D" parallax photos popularized by phone cameras than to a full multi-camera reconstruction, but built and rendered with the same gaussian-splat rendering technique.
Everything here runs on the visitor's own hardware: the depth model is a compact neural network that downloads once and then runs entirely inside the browser, and the splats are drawn by a real-time renderer that sorts every splat back-to-front each frame and blends them with transparency, the same core technique used by full research gaussian-splat viewers. No photo is ever sent to a server — the whole pipeline, from the uploaded picture to the final rendered frame, stays on the device that opened the page.