File formats for keeping photographs for decades

A working format should be small, fast and convenient. A preservation format should be boring, openly documented and widely implemented. Those requirements pull in opposite directions, which is why an archive needs both and has to keep them separate.

What makes a format survivable

Open documentation. The specification is published and complete, so that a reader can be written from it if every existing one disappears.

Independent implementations. Several unrelated programs read it. A format supported by one vendor's product is a format with one point of failure.

No lossy compression in the master. Lossy encoding discards information permanently and compounds through re-saves.

Capacity for embedded metadata. A format that cannot carry the description means the description lives elsewhere and will eventually be separated.

Age. A format that has been widely used for twenty years will probably be readable in twenty more. A format introduced last year has no track record, however good its design.

That last criterion is unfashionable and it is the most reliable predictor. Preservation rewards dullness.

The realistic options

TIFF. Uncompressed or losslessly compressed, openly specified, implemented everywhere, in continuous use for decades, and carries metadata well. Files are large. It remains the default choice for a preservation master and there is no strong argument against it.

PNG. Lossless, openly specified, universally implemented. Designed for screen images rather than photographic capture — it handles colour management and photographic metadata less richly than TIFF — but it is a perfectly sound lossless container and an excellent choice for graphics and screenshots.

JPEG 2000, in its lossless mode. Technically strong, good metadata support, meaningfully smaller than TIFF. Implementation is narrower, which is the reservation: fewer readers means more risk. Used successfully in large institutional archives, and reasonable where storage volume is the binding constraint.

Camera raw. The manufacturer's raw file is the closest thing to the original capture and it is the least survivable, because the format is proprietary, undocumented and specific to a camera model. A manufacturer discontinuing support does not break existing readers, but it does mean no new ones.

DNG. An openly documented raw container. Converting proprietary raw to DNG trades a vendor-specific format for a documented one and loses nothing of substance for most purposes. Where raw capture matters, this is the sensible destination.

JPEG. Lossy, and therefore not a master format — with one important exception. Where the camera produced only a JPEG, that JPEG is the original and must be kept as the master, unaltered. Never re-save it; every save discards more.

Keep the original as well as the master

These are not always the same file, and conflating them loses information.

The original is what came off the camera or the scanner. The preservation master is the archival copy, possibly converted, from which everything else derives. Where conversion happened, keep both — the original as evidence of what was captured, the master as the working archival object.

Derivatives are generated and disposable. A web JPEG can be regenerated from the master at any time, which means it does not need preserving, which keeps the volume of preserved material manageable.

Metadata inside the file

Embed the description in the master. Store it in the catalogue as well, because a file's embedded metadata cannot be searched without opening every file, but embed it regardless.

The reason is separation. Over thirty years a file will be copied between systems, restored from backup, and inherited by people who never saw the catalogue it came from. A file that describes itself survives that; a file that depends on an external index does not.

Where the format cannot hold metadata and a sidecar is used, the pairing has to be enforced by the storage system, because a sidecar left behind during a move produces no error at all.

Format migration

A preservation format will eventually need replacing. The point is to do it deliberately, while the old format is still widely readable, rather than in an emergency when it is not.

That means periodic review: is the format still widely implemented, are new readers still appearing, is there a successor with a real track record. A review every few years is enough, and it is the difference between a planned conversion and a rescue operation.

Conversions must be lossless and verified against the source. A migration that silently degrades everything is worse than no migration.

Conclusion

TIFF for the preservation master, PNG where a lossless screen format fits, JPEG 2000 where volume forces the issue, DNG in place of proprietary raw, and the original camera JPEG kept untouched where that is all there is. Embed the metadata in the file, keep the original alongside the master, treat derivatives as disposable, and review the format choice periodically rather than assuming it. Preservation is a decision you re-take every few years, not one you make once.