Narration that is not a description of the picture

The commonest fault

"Here we see the workers arriving at the factory." Over a shot of workers arriving at a factory.

The viewer has eyes. Telling them what is on screen wastes the only channel you have that can supply something the picture cannot, and it patronises them in the process.

The rule that fixes most narration: words and pictures should do different jobs. The picture shows; the narration supplies what the picture cannot — who these people are, what it costs, what happened next, why it matters, what you cannot see.

Over workers arriving at a factory: "Four years ago there were nine hundred of them. Today there are two hundred and ten."

Write to picture, not before it

The order that works: build the picture edit first, then write narration into the gaps it leaves.

Write the script first and you end up illustrating it — finding a shot for every sentence — and the film becomes a slideshow attached to an essay. Editors call this the radio-with-pictures problem, and it is why so many corporate videos feel like a brochure being read aloud.

The practical method:

1. Cut the sequence with its natural sound.
2. Watch it and note where the viewer will have a question.
3. Write one sentence for each question, in the gap where it arises.
4. Record it, drop it in, and cut the sentences you find you do not need.

Step 4 removes about a third of what you wrote. That third is the sentences that described the picture.

Scratch track first

Record yourself reading the draft on a phone, drop it into the timeline, and edit against it. It costs two minutes and it tells you immediately where the words do not fit the picture — where a sentence needs four seconds and the shot lasts two.

Then rewrite the words to fit the picture, rather than re-cutting the picture to fit the words. The picture was built for a reason.

Replace the scratch with the real recording at the end. Keep the scratch in a muted track until you have; it is the reference for timing.

Writing for the ear

Narration is heard once, at speed, with no ability to go back. That imposes constraints that written prose does not have.

• Short sentences. One idea each. A subordinate clause that works on the page is a place to get lost when heard.
• Say the number in speakable form. "Just under a third" beats "31.7 per cent" unless the precision matters.
• Front-load the subject. "The clinic closed in March" is easier to follow than "In March, following a decision by the district authority, the clinic closed".
• Read it aloud while writing. Anything you stumble over, the listener will stumble over. This is the single most effective revision technique for narration and almost nobody does it.
• Do not write what somebody in the film is about to say. Narration that pre-empts an interview answer removes the reason to hear it.

Recording it for nothing

You need a quiet room, a microphone close to your mouth, and no processing until afterwards.

• Distance: 15 to 20 cm, slightly off-axis so plosives pass beside the capsule rather than into it. A sock or a bit of foam over the mic is a usable pop filter.
• Room: the smallest soft space you have. A cupboard of clothes is genuinely the best free vocal booth in most homes, because clothing absorbs reflections.
• Level: peaks around -6 dBFS, average around -12 to -15.
• Take the time: record each paragraph twice and pick. It is far faster than fixing a bad read.
• Leave five seconds of silence at the top of every recording for noise profiling and room tone.
• Audacity is free and does everything you need for this. So does Resolve's Fairlight page.

A cheap USB microphone at 20 cm beats an expensive one across a room, for the same reason as every other recording in this course: the ratio of direct to reflected sound.

Editing the read

Narration is edited like any other dialogue: remove the breaths that interrupt, keep the ones that are thinking, cut at the gaps between phrases.

One specific: a narration track sounds better with a little more space than feels natural while editing. You are reading along with meaning you already know. Leave a beat after each idea.

Synthetic voices, honestly

Text-to-speech is now good enough to use, and there are three separate questions.

Quality. For neutral informational narration it is usable. For anything carrying emotion, it is still audibly wrong in ways listeners notice without being able to name, because prosody encodes meaning and the model is guessing at meaning.

Disclosure. A synthetic narrator in a factual piece is a fact about the piece. Several jurisdictions now require labelling of synthetic media, the requirements differ, and they are moving.

Whose voice. Cloning a specific person's voice needs that person's explicit, scoped consent — for this project, these uses, this period — and generic stock voices carry their own licence terms about commercial use. Motion, Sound and Music goes into the consent question in detail.

Today

Take any narrated video and transcribe thirty seconds of its narration. Count how many sentences describe what is already visible. That count is the lesson.