Give the voice
a clear purpose.

1. Give the voice a specific job.

A voiceover brief should explain what the audience needs to understand and how the speaker relates to them. “Premium, confident and cinematic” describes a broad mood, but leaves the performer or production team guessing about the actual communication. Is the voice inviting someone in, explaining a product, recalling a memory or making a direct offer?

Start with the sentence the viewer should remember. Then decide what the picture already communicates and what the voice needs to add. If the image shows someone leaving a room, the narration does not necessarily need to describe the departure. It might establish the reason, introduce a contrast or leave space for the image to carry the moment.

Specify whether the voice is an off-screen narrator, a character’s thoughts or dialogue that must correspond to an on-screen person. Those are different production tasks. This guide focuses on off-screen narration; a speaking character also requires decisions about performance, synchronisation and the relationship between voice and face.

Keep the brief independent of the production method at first. An AI-generated film can use a recorded performer, a suitable synthetic voice or no narration. Choose according to the performance and approval needs of the work. The fact that the pictures use AI does not decide what the soundtrack should be.

2. Write for the time the voice actually has.

A twenty-second film rarely needs twenty seconds of continuous speech. The opening may need a beat to establish the scene. A product reveal may benefit from silence. The closing message needs enough room to be understood rather than squeezed into the last available frames.

Mark the windows where speech is useful, then read the script aloud against the picture or an animatic. A word count can help compare drafts, but it cannot prove that a line fits. Pauses, names, emotional emphasis and the chosen delivery all affect duration. Treat a timed reading as an early check, then approve the actual performance in the edit.

If a line needs to be rushed, first ask whether it contains two ideas that could be reduced to one. Speed adjustment can be a finishing choice, but it should not become the default solution to an overloaded script. A calm invitation delivered at a hurried pace changes the meaning of the brief.

The animatic planning guide explains how to map a film before final production. Use that early timeline to test where words belong, while keeping the reading provisional until the voice has been selected.

3. Describe the performance, not an imitation.

Give the team the required language, accent or dialect, intended audience and emotional direction. Avoid relying on a famous person’s name as the entire brief. Describe the qualities that matter: intimacy, clarity, conversational pace, restraint, warmth or a change from uncertainty to confidence.

Use references with a short explanation of what to listen for. One reference might demonstrate relaxed pacing; another might show how a final line receives emphasis. Make clear which quality is relevant so the team is not asked to reproduce an unrelated performance wholesale.

For the fictional Alder Way film, the direction could be “a conversational English-language read, curious rather than grand, with the brand name spoken plainly.” A second candidate could be more energetic, but the comparison should still use the same script and picture. Otherwise a preference for one version may actually be a preference for different wording or editing.

Shortlist voices using the real script’s difficult moments. A demo that sounds appealing on a dramatic monologue may not tell you how a particular name, short invitation or quiet phrase will land. Ask for an appropriately scoped sample where available, then review it against the intended film.

For synthetic speech, record the selected voice and model when known, and distinguish a temporary guide track from approved final audio. For a recorded performer, record the chosen take and direction. In both cases, keep one named person responsible for consolidating feedback so the delivery is not pulled between conflicting instructions.

4. Approve pronunciation before the final mix.

Create a pronunciation list for brand names, product names, abbreviations, locations and technical terms. Include how numbers and symbols should be spoken. “24/7” could be read in more than one way; an acronym might be spoken as a word or as individual letters. Decide what is intended before the recording becomes part of a locked edit.

Provide an approved spoken reference where possible, together with a plain-language note about emphasis. A spelling alone may not communicate the intended sound. Have the brand or language reviewer approve the pronunciation, rather than asking the editor to choose between competing guesses.

ElevenLabs documents pronunciation dictionaries and phonetic or alias approaches, with support varying by model and workflow. Its guidance also recommends testing important pronunciations with the selected voice because results can vary. These are vendor-described controls, not a guarantee that an unreviewed output is correct. Source: ElevenLabs text-to-speech guidance.

Keep the display script separate from any tool-specific spelling used to produce a sound. A phonetic workaround that helps speech generation should not accidentally appear in captions or on-screen copy. Update the pronunciation record when a decision changes, then check the affected lines in the final export.

This is also a reason to avoid treating every voice model as interchangeable. A control that works in one workflow may behave differently or be unsupported in another. Verify the chosen tool’s current documentation and listen to the actual output rather than copying settings from an unrelated tutorial.

5. Review the voice in the film, not only in isolation.

Listen to a clean voice track first to check words, pronunciation, unintended sounds and the delivery. Then listen in the edit with music and effects. A strong standalone read can become difficult to follow if the music occupies the same moment of attention or a sound effect covers the start of a word.

Review the meaning as well as audibility. Does the spoken invitation arrive before the viewer understands what is being offered? Does an emphatic word coincide with a distracting cut? Does the brand name have enough space? These are edit decisions, and they may call for adjusting the picture or soundtrack rather than changing the voice again.

Voquent’s published production workflow includes pacing and dialogue editing, signal processing, file preparation and a final listen-through. That is the provider’s description of its services, not a Nightwolf delivery claim. The useful distinction is that choosing a voice is only one part of preparing a finished soundtrack. Source: Voquent’s workflow.

Ask reviewers to identify the line, time and requested change. “The final invitation feels rushed; allow more space before it” is more useful than “make it sound better.” Separate a performance change from a script change, because replacing words can affect timing, captions and the whole closing sequence.

Use the music brief to coordinate the soundtrack’s priorities, and the caption guide to keep the final words and timing aligned. Review the finished export all the way through; approval of an earlier isolated audio file does not verify the final mix.

6. Hand over a brief that supports delivery.

Before production, agree the intended uses, languages, versions, deadlines and delivery requirements with the provider. Confirm the permitted use of the selected voice through the applicable agreement or service terms. Do not assume that a reference recording authorises voice cloning or that selecting a synthetic voice removes the need to check its permitted uses.

Specify whether the editor needs a clean voice file, separate line pickups, an unprocessed recording or a finished mix. Ask for the format and technical settings required by the actual edit and delivery pipeline rather than treating one export preset as universal. Keep file names tied to the script version, language and take.

A useful voiceover brief gives the production team room to interpret the performance while making the non-negotiable details clear. It protects the message, the timing and the name the audience needs to remember. The result should be judged by how well the voice serves the film, not by how impressive it sounds in a demo.

Sources checked 23 October 2026. Alder Way and its script are fictional; timings are proposed and no recording or generation has been performed. ElevenLabs capabilities and Voquent’s workflow are attributed to their respective providers. No claim is made that Nightwolf used either service. Explore images are independent visual references.