On delivery: four aspect ratios from one master

The least discussed part of the job is the part that eats the most Fridays: delivery. You finish a film, everyone is happy, and then you discover the client needs it landscape for the site, vertical for the phone, square for the feed, and to broadcast spec for the network, each with its own safe areas, loudness target, and file-naming convention. Do that by hand across a campaign and you will lose a day. Do it by hand across a season and you will lose your mind.

So we built the delivery step to run from one master. We finish and grade a single 6K source, one place where colour, sound, and timing live, and everything downstream is a derivation of it rather than a fresh edit. One master means one source of truth: fix a shot once and every deliverable inherits the fix, instead of chasing the same correction through five timelines and missing one.

The derivation itself is a single deliberate pass. Crops for each aspect ratio, a loudness normalise to the right target for each platform, the codec and container each destination actually wants, and file names that match the delivery sheet exactly so nobody is renaming things at 6pm. It runs while we do something else and comes back a set of files that are correct by construction. There is a real satisfaction in watching a two-hour manual chore become a progress bar.

But a crop is not a reframe, and this is where we stop trusting the pass. Taking a landscape master and centre-cutting it to vertical will, sooner or later, put the subject half out of frame or leave a head touching the top edge. The mathematics is fine; the composition is not. So the automated crop is a starting position, never a final one. Every vertical and square deliverable gets a human reframe: a person moving the crop shot by shot so the subject sits where a subject should sit. The machine gives us a draft in seconds; the eye makes it a photograph again.

The same caution applies to sound. Automated loudness normalisation will hit the number, and hitting the number is genuinely useful because every platform has a different one and getting it wrong gets you turned down or rejected. But a mix that measures correct can still feel wrong, because a normalise does not know that the quiet passage is supposed to be quiet. So we let the pass hit the target and then let a person confirm it still feels like the film. Measured-correct and felt-right are two different tests, and we run both.

What we have really automated is the tax, not the craft. The reformatting, the transcoding, the naming, the loudness maths: the parts that are the same on every job and reward nobody for doing them slowly. What is left for a person is the part that was always the point: making sure the vertical cut is composed, the square holds, and the mix still lands. One master in, four honest deliverables out, and the Friday back.