Assembly and editing: which passes to hand over

1The useful dividing line

Editing is not one job, it is about a dozen, and they are not equally suited to being handed over. The line that has held up best is this:

Hand over the passes that are mechanical and reversible. Keep the ones that are taste and irreversible.

Mechanical means there is a correct answer that doesn't depend on how you feel about the piece. Reversible means if it gets it wrong, you can see that it's wrong and undo it in seconds. Almost everything that goes badly in AI-assisted editing is a pass that was neither.

2The passes worth handing over

Transcription-driven cutting. The tool transcribes your footage and you edit by editing the text — delete a sentence in the transcript, the video cuts. This is the single largest time saving in the category and it is barely "AI" in the sense people mean. It is mechanical (the words are the words) and completely reversible.

Filler and silence removal. Strips um, uh, false starts and dead air. Enormously effective, and the one place to be careful: it will also strip the pauses you meant. Always run it with an undo pass — see below.

Rough captions. Near-universal now, and near-universally required, since a large share of short-form is watched without sound. Machine captions get most words right and reliably fluff two categories: brand names and numbers. Both are the words a client will notice. Read every caption before it ships.

Vertical reframing. Taking a wider recording and tracking the subject into a 9:16 crop. Good when there's one subject and it's a face. Poor when two things matter in the frame at once, which is exactly the demo/review case where the product and the reaction are both the point.

Loudness levelling. Mechanical, standardised, and a genuine improvement over most people doing it by ear.

Shot logging and search. Pointing something at four hours of footage and asking which clips contain the pan is a real and undramatic use of the technology, and it saves more time on a busy week than anything flashier.

3The passes to keep

Which take. Two takes of the same line, both clean, one of which is alive. No tool can tell you which, because the criterion isn't in the file. This is the decision your client is actually paying for.

Pacing and where to hold. How long to sit on a beat before cutting is a feel question, and it's where the difference between competent and good lives. The tools' instinct is to remove dead air; sometimes the dead air is the joke.

Comic timing. Same reason, more acutely. Timing is the withholding of the cut, and automated tools are built to not withhold.

The order of the argument. Whether the result goes before the process or after it changes the entire piece. That's a script decision that shows up in the edit, and it's yours.

Anything that changes meaning. Tightening a client quote until it says something they didn't say is the fastest route to losing an account, and it is extremely easy to do by accident with transcript editing. Read the trimmed version back as a sentence. If it now claims more than the original, put it back.

4The undo pass

One habit, and it costs ninety seconds.

After any automated pass, watch the result once, all the way through, without touching anything. Not scrubbing. Not with the timeline open. Watch it like a viewer.

You are looking for exactly three things:

1. Removed pauses that were doing work. Filler removal is the usual culprit.
2. Reframes that cropped out the thing the shot was about. Usually hands or the product.
3. Captions on brand names and numbers.

Those three account for the overwhelming majority of what automated editing gets wrong, and all three are invisible if you only ever look at the timeline.

5Working in a way that survives the client's second round

The revision request always comes, and it is almost always small. The difference between a fifteen-minute change and a two-hour one is decided before you start cutting.

• Keep the original recording untouched. Every automated pass output goes to a copy. Sounds obvious; gets skipped under time pressure, and it is the reason people re-shoot.
• Keep the transcript. If you cut from text, the text is your edit decision list. A client asking to "put the bit about the handle back" is a text search, not an archaeology dig.
• Export a captions file separately where you can, rather than only burning captions into the picture. Burned-in captions mean a typo in a brand name is a re-render.
• Name files so the version is legible without opening them. pan-hook-v3-clientnotes beats final_FINAL_2. Module 5 turns this into an actual system; for now, just be able to tell your own files apart.

6What this does and doesn't do to your rate

Worth being blunt, since Module 0A gave you a rate card and this section just made you faster.

Getting faster does not automatically mean charging less. UGC is priced per deliverable, not per hour — that asymmetry was the whole argument in Module 1. What speed changes is how many deliverables you can accept, and how much of your day goes on the parts that were never the interesting bit.

What it does change is what you're selling. If you're competing on being cheap per video, this makes that race faster and no more winnable. If you're competing on judgement — the right take, the beat held two seconds too long, the line that didn't get cut — the tools make more room for exactly that.

7Exercise 2.4 — One undo pass

Take any piece of footage you already have. Sixty seconds is plenty.

1. Run a filler-and-silence removal pass.
2. Watch it once, straight through, without touching the timeline.
3. Write down every place the automated pass removed something you wanted.
4. Put those back by hand.

Count them. That number is your personal calibration for how much to trust the pass — and it will be different for interview footage than for scripted delivery, which is itself the lesson.