How to Brief Captions
for a Brand Film.

1. Start with what the audience needs to understand.

When commissioning an AI brand film, include captions in the delivery brief before the edit is finished. The useful question is not simply whether words should appear on screen. It is which information a viewer would miss without the soundtrack, and how that information will remain clear alongside the picture.

W3C describes captions as synchronised text for speech and relevant non-speech audio. It distinguishes captions viewers can switch on from open captions that remain visible. Its guidance also warns that automatic output needs checking for accuracy. These principles apply regardless of whether the picture or voice was generated, filmed or recorded. Source: W3C captions guidance.

Write a short inventory of the actual film: narration, dialogue, speaker changes, music and any sounds that carry the story. A doorbell that causes someone to turn is different from an incidental room tone. A voice naming a product is different from a headline that appears only in the image. Identify the job of each element before deciding what belongs in a caption track.

For a film with no speech, do not invent dialogue to fill the screen. Assess whether the audio communicates anything essential. If there is no audio at all, the caption task is different from a spoken film. Ask the receiving team to define the accessibility requirements for the placement; a caption file alone is not a complete accessibility review.

2. Specify the caption deliverables, not just the look.

A buyer should be able to point to the quotation and tell what will arrive. Ask for the language, the matching film version, whether text is permanently in the image, and which separate caption format the destination accepts. Keep a clean picture master when the agreed workflow needs one.

YouTube documents uploading caption files with timing and offers additional creation and synchronisation options. That is a reason to confirm the destination workflow rather than assuming that an export with text burned into it is the only acceptable handover. Check its supported file types for the actual upload. Source: YouTube caption upload guidance.

For a website, involve whoever implements the player. For a social placement, involve whoever uploads the asset. Ask who owns the final preview and who corrects a mistake discovered after upload. Include language versions and re-edit allowances in the production scope so a small delivery detail does not become an unplanned new workstream.

3. Check the words against the final sound.

Use the approved script as a starting point, then listen to the actual final voice. An edited sentence, a shortened pause or an alternate take can make the script an inaccurate record. Generated speech also needs this check: a supplied text prompt is not evidence that every name, number or phrase was spoken exactly as intended.

Give the caption editor a vocabulary sheet containing the approved spelling of brand and product names, unusual terms and speaker names when needed. Ask a reviewer to check figures, negations and claims carefully. A short missing word can reverse a statement; a familiar-looking product name can still be wrong.

Keep expressive text and captions separate in the review. An animated brand phrase may paraphrase the voice for emphasis, while the caption track needs to convey the soundtrack faithfully. If both appear at once, decide which carries each piece of information and check that they do not contradict one another.

In a fictional desk-lamp film, the voice might say, “A little light, wherever you work.” A large on-screen headline reading “Find your focus” is a separate creative message. The caption should not silently substitute that headline for the spoken sentence. This is an invented example, not a Nightwolf commission or a tested asset.

4. Review placement through the whole shot.

Leave room for captions while planning the image, particularly when a product detail, face or demonstration occupies the bottom of the frame. One approved still cannot prove that text remains clear as the camera moves. Review every passage against the actual sequence, including bright areas, shadows and scene changes.

For the fictional lamp film, reserve a caption zone in the composition sketch, then check it again in the edit. If a hand enters that zone to operate the lamp, either adjust the shot or choose another readable treatment. Do not solve the conflict by making the words so small that the design only works on a large monitor.

Break a sentence into meaningful phrases. Watch at ordinary playback speed and ask whether a viewer can read the words while following the image. If a fast voice forces an unreasonable density of text, discuss the pacing of the voice and edit; caption styling cannot solve every timing problem.

Recheck each aspect ratio separately. A landscape composition with spare space can become crowded when adapted vertically. Placement also needs to account for the destination's interface. Use its current preview instead of treating a fixed margin copied from an older project as universal. See our film adaptation guide for the broader framing decision.

5. Treat sound cues and language versions deliberately.

Sound descriptions should help someone follow the film, not list every layer in the mix. In our fictional lamp film, a switch click might mark the exact moment the light turns on. If that cue matters to the sequence, note it for caption review. Do not add a cue to a soundtrack that does not contain it, and do not describe a mood simply because the picture suggests one.

For another language, agree who translates and who approves the result. Provide context and approved terminology, not just isolated caption lines. A change in sentence length may require new line breaks, timing or layout. Review the translated version in motion rather than approving the text in a spreadsheet and assuming the film is finished.

Retain the relationship between each caption file and its picture export. If a new edit removes the opening second, an older timed file may no longer fit. Name the picture version in the handover record and describe what changed. This is especially useful when several lengths and language versions are in production together.

For the creative sound plan itself, our 15-second sound-design example separates the purpose of each layer. The caption brief is a different document: it records what a viewer needs to understand from the finished audio.

6. Sign off the delivered version in its destination.

Review captions twice: once with the sound to check accuracy and synchronisation, then without it to assess whether the audio information remains understandable. These passes answer different questions. Neither replaces checking the upload in the player or app where the audience will see it.

• Words: final speech, names and figures match the approved audio.
• Meaning: relevant speaker changes and sound cues are included accurately.
• Timing: text appears with the relevant moment and remains readable.
• Image: captions avoid obscuring essential action, copy or product details.
• Variants: each language and aspect ratio has its own review.
• Files: the caption file is paired with the exact picture version.
• Destination: the actual upload has been previewed with the intended caption behaviour.

Record the reviewer, version and any unresolved issue. If the wording is approved but the destination preview is still pending, say so. Avoid treating “captions supplied” as proof that the final audience experience has been checked.

For a Nightwolf brief, bring the intended placements, languages and accessibility requirements alongside the creative idea. Our film and post-production capabilities provide the wider production context. Clear caption decisions protect the message as it moves from an edit to a usable delivery.

Sources checked 12 October 2026. The lamp example is fictional and has not been produced or tested. Explore images are independent discussion references, not a single campaign, an audio transcript or before-and-after evidence. No new media was generated.