Audio Description (AD) makes a video accessible to blind and partially sighted viewers. A narrator describes the important things you can only see (actions, settings, on-screen text) in short sentences placed in the natural pauses between dialogue. CAMB.AI can write the descriptions for you, place them in the gaps and voice them.
Two ways to create Audio Description
Video Narration (left sidebar): upload a video and get Audio Description in its language. Upload your own AD script, or let CAMB.AI write the descriptions for you.
Narration mode in a dubbing project: add Audio Description to a video you're dubbing, in the original and the dubbed languages. You switch to it from the mode switcher in the dubbing editor, like Subtitles mode.
Way 1: Video Narration
In the left sidebar, under Projects, click Video Narration.
Upload your video under File, or paste a Link (a Google Drive link must be shared publicly). Supported formats: .mp4, .mov, .mp3, .wav, .flac and .aac, with no limit on size or length.
Choose the Source Language and click Select Narrator voice to pick the voice that reads the descriptions. See What is the Voice Library?
On the right, under Settings:
Project Name: a name you'll recognise later.
Speech Model: MARS8-Instruct is the default, so you can direct the pace and tone with tags (see the tip below).
Script (Optional): already have a written AD script? Upload it as .txt or .docx. No script? CAMB.AI writes the descriptions for you from the video. A script is required if you upload an audio file instead of a video.
Project Setup (optional): choose a folder, add a description and apply keywords.
Click Begin Narration. The project opens in the editor when it's ready; continue with Review the descriptions and generate audio below.
Way 2: Narration mode in a dubbing project
Step 1: Add a narrator
The descriptions are read by a speaker in your project, so you need one with a voice selected. The original speakers use Voice From Original Media and can't be the narrator. If there's no suitable speaker, Narration mode shows "Only speakers with cloned voices can be used for audio description narrator. Please add one from speaker settings."
Open the dubbing project. Next to the project name, click the mode switcher (it shows Dubbing mode) and choose Narration mode.
If you see the message above, click Add speaker. The Speakers panel opens.
Click +, name the speaker (for example AD Narrator), set the gender and languages, and pick a voice: a voice from the Voice Library or one of your cloned voices. Click Confirm. See Speaker Settings and How to create a custom voice.
Step 2: Generate the scene and audio descriptions
Choose Narration mode again. The Generate Scene and Audio Descriptions window opens.
Description: describe the video (what it is, who is in it, the setting). The more detail you add, the more accurate the descriptions.
Check Speakers Names: the descriptions use your speaker names, so make sure they're correct (for example real names instead of "Male guest in black blazer").
Select Narrator: choose the speaker you added.
Click Generate. This runs in the background (you'll see Generating audio description under Active tasks, top-right) and can take a few minutes. You can keep working meanwhile.
Step 3: Open the descriptions
When the task is finished, choose Narration mode again and click Confirm in the Switch to Narration Mode window. It loads your narration transcripts and settings; it's a reversible change, and you can switch back to Dubbing mode at any time.
Review the descriptions and generate audio
The descriptions open in the editor, one line per description, each assigned to your narrator and placed in a gap between dialogue (shown in green on the timeline). The banner reads "Read through the text and make sure everything looks right before generating audio."
Refine the text: click a line to edit it so each description is accurate and fits its gap. You can also move or resize a line on the timeline, or add, split, merge and delete lines. See How do I edit my Audio Description in the editor? and The dialogue toolbar and icons.
Generate audio: click Looks good, generate audio, then Confirm in the Regenerate All Audio? window. This can take a while for long videos.
Listen and fix timing: the banner now reads "Listen through the audio and adjust timing, speed or speakers." Play the video through and check that each description matches the picture and sits in a natural pause. A speed badge (for example 1.44x) and a count such as "3 dialogues with speedups / gaps" show descriptions that are too long for their gap: shorten the text, widen the slot if there's room, or change the pacing tag, then re-generate that line. See How to re-generate dialogues after editing and gaps and fast dialogues.
Audio Description in a dubbed language
In a dubbing project, the descriptions are created in the original language first. Open the dubbed language tab (for example German (GE)) and click Translate to translate them.
The dubbed dialogue has different timings from the original, so some descriptions may now overlap it. Those lines are flagged (for example "4 dialogues with errors") with the blocker "Audio descriptions cannot overlap with dialogues." Shorten, move or delete those lines until the errors are gone, then generate the audio for that language.
Tip: control the pace with MARS8-Instruct
When a gap is tight and a line sounds rushed or too slow, add a pacing instruction such as [fast paced] or [concise, brisk pace] to that line and re-generate. See the Emotion Tag Gradation Guide and Which speech model should I choose?
Export
When every description sounds right, click Sounds good, generate final mix. When the final mix is ready, click Export (top-right) to download the video with the Audio Description, the audio, or subtitles. See How to download output files.
What should I describe?
Actions and body language: movements, facial expressions, gestures and where people are, when they drive the story. For example "Carries his box to the table." or "She frowns."
Settings and scene changes: the location, time of day, lighting and transitions.
On-screen text and graphics: opening titles, name and role captions (lower thirds), scoreboards, prices and other key graphics.
Appearance and costumes: clothing, props or details that help identify a character or set the tone. For example "Adam wears a bold Hawaiian shirt."
How should descriptions be written?
Present tense: "Stephen steps to a contestant", not "Stephen stepped".
Objective and neutral: describe what is visible. Don't explain motives, judge emotions or add literary flourishes.
No spoilers: never reveal a plot twist, punchline or reveal before it happens on screen. A split-second lead-in is fine when dialogue starts right after a quick visual cue.
Concise and specific: use strong verbs and exact nouns. Skip filler such as "We see…" or "On screen…". Spell out numbers and abbreviations the way they should be spoken.
What timing standards should I follow?
Never talk over dialogue, commentary, lyrics or sound effects that matter to the story. Broadcast Audio Description usually follows these benchmarks:
Speech rate: 140–160 words per minute (about 2.5–3 words or 14–16 characters per second).
Minimum gap: a pause of at least 1.5–2 seconds between dialogue lines to fit a short description without rushing.
Cue length: 1–2 short sentences per description, roughly 40–120 characters depending on the gap.
Safety buffer: leave about 0.25–0.5 seconds before and after dialogue so the description never clips the speech.
ℹ️ Use these numbers as a guide, not hard limits. AI voices adapt their pace to each line, so check each description by ear rather than by word count alone.
Checklist before you export
No description overlaps dialogue, lyrics or important sound effects (in every language).
Each description is in the present tense, objective and free of spoilers.
On-screen text, titles and name captions are read out.
Names are correct and pronounced correctly. If not, see How to make your audio sound good.
You've listened to the whole video at least once from start to finish.







