Content or data? What we keep when we keep a film

“We keep the film” can mean one terabyte or ninety. The difference is not one of quality. It is a choice between preserving the content, the work in the forms in which it can be shown again, and preserving the data: everything the work was made from.

Where a rule gives preservation a shape, as in France, keeping a film stops being a fear and becomes a line in a budget, and a market for preservation service providers. We set out that argument in our Manifesto for the Memory of Digital Cinema, presented in Berlin in February 2026. This paper takes the next step. Once someone is committed to keeping a film, what are they committed to keeping, at what cost, and who should carry it?

We put numbers on one two-hour feature, read them against the French technical recommendation, and argue for drawing a line. Content belongs to rights holders. Data, kept for the very long term, is a legitimate mission for public custodians. The levels in between should be decided explicitly, and early.

When a producer, a distributor or an archive says “we keep the film”, the sentence sounds complete. It is not. Behind it sit very different quantities of material, very different costs and very different intentions. To make the question concrete, take one film, two hours long, finished in 4K, and add up what keeping it can mean.

Level 1 · The finished master

The finished master is the film as it was released. We use Apple ProRes 4444 as our example because it is the most widely used mezzanine format in the industry, but the same reasoning applies to DNxHR or any comparable codec. At 4096 × 2160 and 24 frames per second, ProRes 4444 runs at about 1,130 Mbit/s,¹ so two hours come to about one terabyte.

Level 2 · Adding the DCDM

The Digital Cinema Distribution Master (DCDM) is the standard master of digital cinema: uncompressed 4K images, one file per frame.² For the same two hours it adds about nine terabytes. Level 2 totals ten terabytes, ten times level 1.

Level 3 · Adding the source material

Then there is everything the film was made from: camera originals, rushes, and the project files of editing, grading, visual effects and sound. In our experience, an average feature adds around eighty terabytes.³ Level 3 totals about ninety terabytes, roughly ninety times level 1.

France is the clearest reference we know. To obtain production approval from the CNC (Centre national du cinéma et de l’image animée, the French public body that funds and regulates cinema), a film must supply a preservation contract with a technical provider, and for the digital elements that contract refers to a recommendation written with the industry through the CST, the French technical commission for image and sound. The current text is CST-RT-043, revised in May 2025.⁴ A note published with it proposes extending the obligation from films to all audiovisual works, and lengthening the minimum term beyond five years.⁵

For a feature film it asks, in substance, for level 2:

Look again at the three levels. The first two have a boundary. You can list the files, count the terabytes and know exactly what it costs to keep them. They describe the work as it was released, and as it can be released again.

The third has no boundary. Every production creates a different amount: a documentary shot over three years and a studio comedy shot in six weeks have nothing in common at this level. Even the French recommendation, the most complete rule we know, only says “possibly”.

So we would draw a line between the two, and give each side its own task.

When the mission is to preserve the data, spending has no ceiling. Every film adds its ninety terabytes, and every year adds more films. Storage that grows with each new title, with no end date, is not sustainable for any archive or any budget.

This is not a criticism of archives. Some of this material is the very reason archives exist. The point is narrower and, we think, harder to dispute:

Moving from level 1 to level 2 multiplies storage by ten. The honest question is whether a ProRes 4444 master, on its own, preserves the film.

What the evidence says

Apple’s own documentation describes ProRes 4444 as offering visual fidelity “perceptually indistinguishable from the original material”.¹ That is a vendor’s claim, so we looked further. A post-production facility tested ProRes 4444 against uncompressed DPX, measuring colour difference (ΔE), PSNR and SSIM, and found no perceptual difference.⁷ The test has not been peer-reviewed, and we present it as such.

What the audience sees

Consider what reaches a cinema screen. A DCP is compressed to 250 Mbit/s at most.² The same film as a 4K ProRes 4444 master runs at about 1,130 Mbit/s, more than four times as much. A ProRes master can also carry a wider colour gamut and a higher dynamic range than the DCDM. The DCDM is graded for a cinema projector, so its colour and light are bounded by a theatrical presentation.

From level 2 to level 3, storage grows about nine times again. Here there is no quality percentage to measure: the film on screen is already fully preserved. The source material secures something else.

What you pay for is an option: the option to remake the film. A new grade for a new display technology, a new version, a restoration from the original. These are real needs, but in our experience recovering original footage is rare. And when what someone needs is a cut scene or some B-roll for a documentary or a re-release, a mezzanine copy of the rushes may be enough. It costs a fraction of the camera originals.

Content and distribution are not two separate stories. A master you keep but cannot deliver earns nothing back. What you preserve is worth exactly as much as your ability to put it back on a screen: in a cinema, on a broadcast, at a festival.

Put together, the three levels are not a ranking of good, better and best preservation. Each serves a different purpose, and each has its price.