VOLUME
← Learn
Learn · 4 min read

Virtual tour from an iPhone video: how it works

Short answer

A virtual tour from an iPhone video works by pulling a few hundred still frames out of the walkthrough, solving where the phone was for each frame, and training a Gaussian splat that reproduces the room from any angle. You film one slow, steady walk through every room; the pipeline does the rest. The result is a 3D scene a visitor can move through in a browser, not a stitched panorama or a video with hotspots.

VOLUME Virtual

What kind of video do you need?

One continuous walk. Landscape, 1080p is enough, 4K is fine but not necessary. Between 60 seconds and 5 minutes long (Team plans accept up to 8). MP4 or MOV, under 500 MB.

The phone's default Camera app is the right tool. Turn off cinematic mode and any portrait effect; those add synthetic blur that the reconstruction reads as real. Lock exposure if your phone lets you, so the brightness does not pump as you turn toward a window.

Walk slowly. Keep the phone at head height. Cover every room edge to edge, and let the camera see each wall from more than one position. The pipeline can only rebuild what the camera saw, so a corner you never pointed at becomes a hole in the tour.

What happens to the video after upload?

Four stages, all automatic.

  1. Frame extraction. The pipeline samples sharp frames from the video, spaced across time and camera position. Blurry frames from fast pans are skipped where possible.
  2. Camera solve. Structure-from-motion (SfM) matches features between frames and works out the phone's position and orientation for each one. This is the invisible skeleton every 3D reconstruction needs. If the walk was too fast or the room too featureless, this is the stage that fails.
  3. Splat training. Millions of small coloured ellipsoids are placed in space and adjusted until, rendered from each solved camera position, they match the frame the phone actually recorded. This is the bulk of the compute.
  4. Publishing. The finished splat file is stored, given a public URL, and wrapped in a web viewer. The same file is yours to download.

None of these steps involve a person looking at your footage. The job is dispatched to a reconstruction vendor and polled until it finishes.

Why a splat instead of a 360 photo?

A 360 photo tour is a set of panoramas. You can look around from each spot but you cannot step between them; the viewer jumps. A Gaussian splat is volumetric. Move the camera a metre to the left and the scene re-renders correctly from the new position, with real parallax: the chair in front slides across the window behind it.

That is the whole difference. A splat behaves like a room. A panorama behaves like a photo you can spin. The plain-english explainer at What is a Gaussian splat? goes deeper, and Gaussian splats vs 360 photos vs video tours compares them side by side.

What comes back, and how long does it take?

You get three things: a public tour URL, an iframe embed snippet, and the reconstructed file to download. The URL works anywhere a link works. The embed drops into any page that accepts an iframe.

Turnaround is typically under an hour from upload. That is a target, not a promise. Longer videos take longer, and the vendor queue can back up. Read what "under an hour" means before you plan a listing launch around it. If reconstruction fails, you are not charged and the attempt does not count against a monthly quota.

What goes wrong most often?

Three things, in order.

  • Walking too fast. Motion blur kills feature matching. If you think you are walking slowly, walk slower.
  • Uneven light. Blinds open in one room and closed in the next makes the camera re-expose at every doorway. The reconstruction inherits every one of those jumps.
  • Reflective and featureless surfaces. Mirrors, floor-to-ceiling glass, and bare white walls give the camera solver nothing to lock on to.

Each of these has a fix in the shot guide and the longer shooting for reconstruction companion. If a tour has already come out soft, see why your 3D tour came out blurry.

What VOLUME does here

VOLUME is the upload-and-wait version of this process. You film the walk on whatever phone you have, upload the MP4 or MOV in the dashboard, and the reconstruction is dispatched to a third-party Gaussian-splat pipeline. When it finishes you get a tour URL, an iframe snippet, and the file. No app to install, no hardware to buy, no photographer to book. Your first tour is free; after that a single tour is $79 with 12 months of hosting included.

Questions people ask

Do I need an iPhone Pro with LiDAR?
No. The reconstruction is built from the video frames, not from a depth sensor. Any recent iPhone or Android phone that records steady 1080p video works.
Can I use an existing video I already shot?
If it is a slow, continuous walkthrough that covers every room, yes. A highlight reel with cuts, music overlays, or fast pans usually will not reconstruct.
Does someone at VOLUME watch my video?
No. The job is dispatched to the reconstruction vendor automatically. The founder only looks at a tour if you open a support request about it.

Try it on your next listing.

Your first tour is free, no card at signup. After that a single tour is $79. If a usable walkthrough can't be reconstructed, you're never charged.