Voiceover or dialogue: give your AI film the right voice

Listen to what the image cannot say

Imagine a fictional night-time museum opening. A hand unlocks a door, a visitor pauses before a sculpture, and a light reveals its outline. I would not add a voice saying that a door opens. The useful words might explain why this evening matters: a chance to see a familiar place differently. That gives the voice a job instead of filling every quiet moment.

I first write the thought I want someone to take away, then watch the planned images without speaking. What is already clear? What would a newcomer misunderstand? If the missing piece is an idea, narration may help. If it is a relationship between two people, a short exchange may serve it. If it is merely the name of the event, a readable title may do enough.

Compare three routes before choosing a voice

A voice outside the picture lets me change the wording without matching every syllable to a visible mouth. A speaking character adds a performance to judge: delivery, pauses, expression and movement must belong together. A film without speech places more responsibility on the pictures and titles. I discuss those trade-offs before building the full sequence.

For the museum example, the same invitation could work as a calm narrator over the rooms, a conversation between two visitors, or a sequence led by footsteps and a final event card. I would make a small rough version of the strongest two options. This is a creative comparison using our own material, not a claim that one tool or style wins for every brand.

Leave room for a sentence to land

A written script can fit on one small page and still crowd a short film. I read it aloud, including names and numbers, while timing the whole take. A pause that lets us notice the sculpture is part of the film, not wasted time. If the line is too long, I usually remove a thought before asking the speaker to hurry.

The timing worksheet uses your word count, your chosen speaking speed and your allowance for pauses. It is a planning estimate. It does not hear your voice, judge the performance or predict lip synchronisation. A recorded take can be slower or faster; use its measured duration when you have it. A translated line also deserves a fresh read rather than the timing of the English script.

• Read the names exactly as they should sound
• Mark where the picture needs attention without words
• Time the complete take, including breaths
• Keep a small margin for the opening and ending

Make one difficult passage before the whole film

I test the line that contains the brand name, an emotional turn or a difficult action. With a narrator, I listen for pronunciation and whether the emphasis changes the meaning. With a character in the picture, I also watch the mouth and the face during pauses. A convincing first word does not prove that the whole exchange works.

Then I place the test in a rough edit. The voice may sound lovely on its own but compete with a loud musical phrase, or arrive before the viewer can read the event card. I adjust the relationship between these elements in the edit. There is no universal volume setting that can replace listening to the actual mix on ordinary speakers.

Give every language its own performance

For a film from SOLONSKY Films, I agree the spoken text, voice direction and intended languages with you. A translation should sound like something a person would say in that language. I may shorten a clause or move a pause while keeping the meaning. Any change to a factual claim needs your approval; natural phrasing is not permission to invent a new promise.

I ask the reviewer to compare the final spoken words with the approved text, check captions against the actual take, and watch the ending on a phone. Keep the voice track separate from music and effects if later language versions are part of the agreed delivery. This leaves a practical way to replace the narration without starting the entire sound edit again.