Turnaround: what "under an hour" means and when it's longer

Five steps, in order.

1. Upload. Your browser sends the video straight to storage. A 300 MB file on a decent connection takes a minute or two. This is the only step that depends on you.
2. Dispatch. The server streams the video to the reconstruction vendor and starts a job. Seconds, with automatic retries if the hand-off hiccups.
3. Queue. The vendor's pipeline picks up the job when a worker is free. Usually quick; sometimes not, if many jobs arrived at once.
4. Reconstruction. Frame extraction, camera solve, splat training. This is the bulk of the time and scales with the length of the video.
5. Publish. The finished file is mirrored to VOLUME's storage, the poster is generated, and the tour page goes live. A minute or so.

The "under an hour" figure describes the total for a typical 2–4 minute walkthrough under normal pipeline load.

Because the slowest step runs on infrastructure VOLUME does not control. The reconstruction vendor is a third party with its own queue, its own capacity, and its own bad days. VOLUME can retry, redispatch, and watch, but it cannot make a GPU appear.

Saying "typically under an hour" is accurate. Promising a fixed hour would be a commitment VOLUME cannot keep on a bad afternoon, so it does not make one. A contracted turnaround SLA is part of the Custom tier that is in development for brokerages; it is not for sale yet and the pricing page says so.

• Video length. A 5-minute walk has roughly twice the frames of a 2.5-minute one, and training time grows with it. Team plans allow 8 minutes; expect those to run longer.
• Pipeline load. Jobs are processed as workers free up. Evenings and the start of the work week tend to be busier.
• Difficult footage. A walk with a lot of motion blur, dark rooms, or reflective surfaces gives the solver more to reconcile and can slow training or cause a retry.
• Queue stalls. Occasionally a job sits in the vendor's queue without starting. VOLUME's reconciler redispatches these automatically after a set interval, up to a limit, before marking the job failed.

None of these are things you can see from the outside, which is why the dashboard shows the current state rather than an estimate it cannot stand behind.

Nothing, mostly. The reconstruction screen polls while the tab is open; closing the tab does not cancel the job, and a server-side cron reconciles status in the background. You will see the tour appear in the dashboard when it is done.

If you are planning a listing launch, upload the day before. Treat the hour as the usual case and give yourself a buffer for the unusual one. If a tour is not back after a few hours, open a support message from the dashboard; the founder reads them.

You are not charged, and the attempt does not count against a monthly quota. The dashboard shows the failure and, where the pipeline reported one, the reason. The common reasons are the ones the shot guide exists to prevent: too fast, too dark, not enough coverage. Jobs that run far past the expected time are failed automatically after a cut-off so they do not hang forever.

Reshoot, upload again, and the clock restarts. Why your 3D tour came out blurry covers how to read the failure and fix the walk.