Why a three-minute video takes all evening.
The arithmetic is insulting. Three minutes of finished video is maybe 450 spoken words. You can say 450 words in one breath-managed sitting. And yet you sit down at eight and you are still there at eleven, on take nineteen, having lost the ability to pronounce your own company name.
The time does not go where people think it goes. It is almost never the words. It is a small number of specific, unglamorous mechanisms, and each one has a fix. Here are the eight I keep hitting.
1. You are still writing while you record
This is the biggest one and the least discussed. You start a take, get four sentences in, and think "actually that's not the right order". So you stop. You edit. You start again. You are now doing two jobs at once, and one of them requires you to be calm in front of a lens while the other requires you to be critical of your own text.
Fix: finish the script before the camera is on. Not "mostly finish". Finish, then read it out loud once end to end with no camera running. Every edit you would have made mid-take, you make there, for free.
2. You restart from the top every time
You fluff a word ninety seconds in, and something in your brain says "ruined, start over". So the first thirty seconds get recorded nineteen times and the last thirty get recorded twice, badly, by an exhausted person.
Fix: record in blocks. Decide your cut points in advance — usually the section boundaries — and treat each block as its own take. When you fluff, back up to the last block boundary, not to the top. Clap or say a marker word so the cut is trivial to find later. This one change does more for your evening than any piece of equipment.
3. The pace is not yours
If you are reading from a fixed-speed scroll, you are spending the whole take managing a race condition between your mouth and a timer. Chase it and you fluff. Fall behind and you stall. Either way, that is a retake, and it is a retake caused by the tool rather than by you.
This is the failure mode I know best, and I wrote a whole piece on why a fixed scroll makes you sound like you are reading. The short version: there is no correct scroll speed, because your own speed changes several times per paragraph.
Fix: stop reading against a clock. Either learn the block well enough to say it without prompting, or use a prompter whose scroll follows what you have actually said rather than a timer. If the pace mismatch is what is eating your takes, that is the specific problem I built my own app to remove, and seven days of full access is enough to count your takes before and after and see whether the number moves.
4. Your setup is not repeatable
Chair height drifted. You moved the laptop to check something and never put it back. The lamp got nudged. Between take four and take twelve, your framing has quietly changed, and now the good take does not intercut with the other good take.
Fix: mark it. Tape on the desk where the laptop sits, tape on the floor where the chair goes, and a screenshot of your framing pinned somewhere you can see. Ten seconds of setup discipline buys back a whole category of "the good take doesn't match" pain.
5. You are watching yourself while performing
The self-view is a trap. You are simultaneously trying to speak naturally and monitoring your own face for whether you look like an idiot. Nobody can do both. The monitoring wins, and it makes you stiff, which makes you retake, which makes you monitor harder.
Fix: turn the preview off, or cover it. Check framing once before you roll, then hide it. If you need a reassurance signal, keep a small recording indicator visible and nothing else.
6. You discover the audio problem afterwards
This is the expensive one, because you do not lose a take, you lose the whole session. The gain was wrong. The fridge kicked in. The wrong input was selected. A notification chimed. You find out during editing, and there is no fixing it in post beyond a certain point.
Fix: a ten-second test recording every single time, played back on headphones before the real take. And put the machine in Do Not Disturb — silencing notifications is a two-second action that has saved me more takes than any microphone upgrade.
7. You are operating slides or a screen mid-take
If you are also driving Keynote, PowerPoint or a demo, you are now a presenter and a stage manager at once. Every slide advance is a moment where your hand leaves the frame, your eyes leave the lens, and your sentence goes flat while you look for the arrow key.
Fix: get the slide changes off your hands. Either record the slides as a separate pass and cut them in later, or drive them from the script itself so the advance happens where you wrote it rather than where you remembered it. Separating the two jobs is the point, however you do it.
8. You have no stopping rule
Without a rule, "good enough" never arrives, because take nineteen is being judged by someone who has now heard the same paragraph nineteen times and finds every version of it unbearable. That judgement is worthless. You are not evaluating the video any more, you are evaluating your own fatigue.
Fix: decide the rule before you start. Mine is three takes per block. If block four fails three times, the block is written wrong — I go and rewrite it rather than perform it harder. And whatever survives after three attempts is the one that ships. The audience is hearing it once. You are hearing it for the twentieth time. Your ear is not the relevant instrument.
The pattern underneath all eight
Every item on that list is the same mistake wearing a different coat: doing two jobs at the same moment. Writing and performing. Performing and monitoring. Speaking and stage-managing. Reading and racing a timer.
Every fix is the same move too: separate them in time. Write, then read aloud, then set up, then test audio, then perform in blocks, then judge tomorrow. Serialised, a three-minute video is a twenty-five minute job. Done all at once, it is genuinely an evening, and you will not enjoy any part of it.
One more thing worth checking before you start: whether the script is something you are even allowed to put into a cloud tool. If it names a client or covers something unannounced, that question deserves an answer before the camera goes on, not after.
← All posts