Gaussian splats vs 360 photos vs video tours

Video captures one path, at one speed, from one height, chosen by whoever held the camera. The viewer sees exactly that and nothing else.

360 photos capture everything visible from one point, in every direction. A tour is several of these points linked by hotspots. From each point the viewer can look anywhere; between points, nothing was captured.

Gaussian splats capture the volume of the space. The reconstruction solves where millions of small, semi-transparent coloured ellipsoids must sit so that every frame of the source video is reproduced. The result can be rendered from positions the camera never occupied, within reason: step off the filmed path by a metre and it still holds; walk into a room the camera never entered and it falls apart.

• Video: play, pause, scrub. The viewer cannot look left if the camera looked right.
• 360: rotate the view, zoom, click a hotspot to jump to the next panorama. The jump is a cut, not a walk. Dollhouse views in some products are a separate rendered model, not the panoramas themselves.
• Splat: move forward, back, sideways, turn, look up and down, continuously. Parallax is real: near objects move faster than far ones. This is the thing that makes a splat read as "being there" rather than "looking at a picture of there."

A buyer deciding whether a sofa fits between the window and the door can answer that in a splat by walking to the spot. In a 360 they estimate. In a video they wait and hope the camera went there.

Video: a phone, a minute, and nothing else. Edit if you like. Upload to YouTube or the listing. This is why every listing has one and few are useful.

360: a 360 camera (Ricoh Theta, Insta360, or similar; roughly a few hundred dollars), a tripod, and a hosting platform to link the panoramas and serve the viewer. Each panorama is one placement of the tripod. A home takes 15–40 placements. Some products let a phone on a rotating mount stand in for the camera.

Splat: a phone video, shot slowly, and a reconstruction pipeline to turn it into the file. No extra hardware. The constraint moves from equipment to technique: the walk has to be slow, lit, and complete. Reconstruction is compute-heavy and takes time; VOLUME targets under an hour and does not guarantee it.

Video breaks on interactivity. It also goes stale fast; a 2-minute clip with music reads as marketing and gets skipped.

360 breaks between the points. Thin spaces like hallways and stairs need many placements or they feel like teleporting. Mirrors and moving people are stitched in as artifacts. The file set is large and the viewer usually needs the host's player.

Splats break on coverage and on surfaces. A room the camera did not walk is not in the tour. Mirrors, glass, and glossy screens confuse the solver. Thin geometry like railings and plants can look fuzzy up close. The file is a single asset (commonly a .ply or compressed .spz) that needs a WebGL viewer; most browsers from the last few years have one built in, so this is a smaller problem than it was. Splats also have no built-in measurements or floor plans; those are separate products.

Photos first, always; they get the click. Then one interactive format, not three. If the property's selling point is layout and flow, a splat shows it. If it is a single dramatic room, a 360 of that room may be enough. If you have thirty seconds and no intention of doing more, a video is better than nothing.

The format matters less than whether the viewer can answer their question. The Matterport comparison covers the hardware-based 3D option in detail, and the glossary defines each term used here.