Backup for photographers, and what a delivery gallery is not

Every photographer who has been working for a few years has a version of the same story, or knows someone who does: a drive that stopped mounting, a card that corrupted before it was offloaded, a laptop stolen from a car. The photographs were not backed up because backing them up was a task for next week.

This is the short, practical version of what a backup actually is — and, because it is the specific confusion this site sees most, what it is not.

A client gallery is not a backup

Every delivery platform, this one included, is a way of moving photographs to somebody else. That is a different job from keeping them safe, and the two get conflated because both involve a copy of your files sitting on a server.

Three concrete reasons the conflation hurts:

• You deliver finished JPEGs. Your negatives — the RAWs, the catalogue, the edits — are not there.
• Galleries get deleted, expire, or are removed to reclaim storage you are billed for. That is normal delivery hygiene and catastrophic archive behaviour.
• An account is one password reset away from being someone else's, and one billing failure away from being suspended. Neither is a property you want in your only copy.

Use the gallery to deliver. Keep the archive yourself. The two are not competing and should not be the same thing.

The 3-2-1 rule, stated properly

The conventional formulation, and it is conventional because it survives contact with reality:

1. Three copies of the data.
2. On two different kinds of storage.
3. One of them off-site.

The part people get wrong is the counting. The working copy on your machine is copy one. A clone on a drive that sits next to that machine is copy two, and it is on the same desk — so a flood, a theft or a power surge takes both. That is why the third has to be somewhere else, and why “somewhere else” is the requirement rather than “in the cloud” specifically.

A version that a working photographer will actually maintain

The best backup scheme is the one still running in eighteen months. This one has three moving parts and no daily decisions:

• Offload before you sleep — cards to the working drive the same night, always, and do not format the card until the next backup has run. The card is a free extra copy for as long as you leave it alone.
• An automatic local mirror — a second drive that copies the working drive on a schedule you did not have to remember. Manual copying is a scheme that lasts until the first busy week.
• An off-site copy — a cloud backup service, or a drive kept at a family member's house and swapped every month or two. The swapped drive is unglamorous and completely effective.

Note what this costs in attention: one habit at the end of a shoot, and one drive swap every few weeks. Everything else runs without you.

The restore test

This is the section that separates a backup from a belief about a backup. Once, pick a job from six months ago, and restore it from your off-site copy onto a machine that has never held it.

You are checking four things, and each one has failed for somebody:

1. That the files are actually there — backup software silently excluding a folder is common.
2. That they open, and are not truncated or corrupt.
3. That you can find them without the original catalogue, using nothing but folder and file names.
4. That you know your own credentials for the service, without the password manager on the machine that just died.

That last one is the failure that surprises people. An untested backup is a hypothesis, and the moment you need it is the worst possible time to test it.

Filenames are part of the backup

In ten years the catalogue may be unreadable and the folder structure may have been reorganised twice. What survives is filenames. A pattern like 2026-07-18_hana-omar_ceremony_042.jpg sorts chronologically, says what it is, and does not collide with photographs from any other source. DSC_4471.jpg does none of that. There is more on this in export settings for client delivery.

How long to keep it

Storage is cheap enough that the answer for finished work is “indefinitely”. The decision worth making is about the material in between: rejected frames, duplicate brackets, the fourteen near-identical portraits you culled to one.

A defensible rule is to keep every selected RAW and every delivered JPEG forever, and to cull the rest within a month of delivery while you still remember which was which. Doing it later means doing it never, and the volume grows faster than the resolve to sort it.