AI Video & Voice Production: ElevenLabs, Veo, Runway and MoreQuality, accessibility and disclosure · Lesson 13 of 16
Captions, accessibility and quality control
Video lecture
Captions, accessibility and quality control
The narrated lecture is in production
Every chapter is scripted and ready. Browse the chapters and read the full transcript now — the video will appear here when it’s published.
Chapters
Transcript of the narration, chapter by chapter.
0:00 Captions, accessibility and QC
Here's a habit that should change how you edit: on social platforms, lots of people watch with the sound off, at least at first. You've probably done it yourself, on a train, in a waiting room, or late at night next to someone sleeping. If your video doesn't work silently, it doesn't work. And for viewers who are deaf or hard of hearing, captions aren't a preference, they're access. In this lecture you'll learn a captioning workflow, accessibility practices beyond captions, and a quality-control checklist built for AI media.
0:39 Why it's quality
Why is this part of quality, not an optional extra? Three reasons. Accessibility: captions and good narration make your content usable by more people, and accessibility laws and public-sector requirements in several markets reinforce this. Reach: sound-off viewing is common, and non-native speakers rely on captions to follow. And risk: AI media fails in small, specific ways, like a mispronounced name, a lip-sync glitch, or a morphing hand. A structured QC pass is what catches those before your audience does.
1:14 The pre-flight mindset
Here's the mental model for QC. Think of an airline pre-flight checklist. Pilots don't skip it because they're experienced. They use it because experience makes you confident, and confidence makes you skip things. AI media is the same. After fifty videos, you'll feel sure the captions are fine and the pronunciation is right. The checklist is there for the day they aren't. It covers four areas: audio, visuals, content and compliance, and ends with a final phone check.
1:48 Captioning workflow
Let's start with the captioning workflow. First, auto-generate captions in your editor, avatar platform or a transcription tool. Second, correct them, because automatic captions often get names, brand terms, numbers, local words and mixed languages like Urdu and English wrong. Third, style them: high contrast, big enough for mobile, one or two lines, broken at natural phrases, and positioned away from faces and platform buttons. Fourth, check timing. Fifth, caption each language version in its own language. And sixth, upload caption files where platforms support them.
2:25 Why caption files
Why upload a caption file when captions are already burned in? Because a separate file, usually an SRT, lets viewers turn captions on or off, lets platforms index your words, and lets you fix a typo without re-exporting the whole video. An SRT is simple: a number, a start and end time, and one or two lines of text, repeated. On YouTube, for example, you can upload one per language. Burned-in captions are still great for social feeds where autoplay is silent. Often you'll want both.
3:02 Beyond captions
Accessibility goes beyond captions. Describe key visuals in the narration when they carry meaning. Instead of saying as you can see on the chart, sales doubled, say sales doubled from March to June. Avoid rapid flashing effects that can harm people with photosensitive conditions. Keep strong color contrast for on-screen text. Use plain language, which helps everyone, including viewers using translation tools. And add alt text to thumbnails and images in posts where platforms support it. Ask yourself: would this make sense to someone who can't hear it, and to someone who can't see it?
3:43 QC checklist, part 1
Now the QC checklist itself. Audio: every word pronounced correctly, no glitches or sudden volume jumps, a consistent voice across scenes, and licensed music that doesn't overpower speech. Visuals: accurate lip-sync, no distorted hands, faces, text or morphing objects, correct brand colors and fonts, and no unintended recognizable people, logos or sensitive locations. Content: facts, prices and claims match approved sources, localized versions reviewed by native speakers, cultural sensitivities considered, and generated visuals illustrate but never prove.
4:16 QC checklist, part 2
And compliance. Consent on file for any cloned voice or custom avatar. Platform AI-content labels applied where required. Sponsorship disclosures present where applicable. Captions accurate in each language. Then the final line, which is my favorite: watched in full, on a phone, with sound on and with sound off. Why a phone with sound off? Because that's how many short-form viewers watch. This one pass reveals captions hidden under platform buttons, text too small to read, and whether the video still makes sense silently.
4:53 Example 1: Dubai fitness promo
First example, a simple one. A Dubai fitness studio's AI-dubbed Arabic promo passes the audio check. Then the sound-off phone test reveals two problems. The captions are still in English, and they sit under the like and comment buttons. The team regenerates Arabic captions, corrects two misspelled class names, moves the captions up, and publishes. That ten-minute check prevented a confusing launch, and it only happened because the checklist said phone, sound off.
5:25 Example 2: UK public-sector training
Second example, a business case. A UK training company that sells to public-sector clients adopts a caption standard: sentence case, two lines, speaker labels when there are two voices, and sound cues like music fades where they matter. They add an accessibility statement to each course page. And their QC checklist gains one line: would this make sense to someone who can't hear it, and to someone who can't see it? That question keeps catching narration that says look at this without describing what this is.
6:02 Watch me do it, part 1
Watch me do a quick automated caption check before the human review. I run the small script from the lesson on my SRT file. It splits the file into blocks, reads each timestamp, and flags three things. More than two lines in a block. Any line over forty-two characters. And any caption shown for less than one second, which is too fast to read. It prints the block numbers that fail. I fix those, rerun until it's clean, and only then hand the video to a reviewer who reads every word.
6:42 Watch me do it, part 2
Then the human pass. My reviewer isn't the person who made the video, because fresh eyes catch what the maker's brain auto-corrects. They watch once without stopping, noting timestamps. Then they go through the checklist section by section. For multilingual releases, each language gets its own native reviewer. I time-box it at roughly twice the video's length plus fixes. A two-minute video gets about five minutes of review. It's the cheapest part of the whole production, and the one that protects everything else.
7:18 Common mistakes
Common mistakes. Trusting auto-captions without reading them. Burned-in captions only, with no caption file where the platform supports one. QC by the same person who made the video when a fresh pair of eyes is available. Leaving English captions on a dubbed version. And narration that relies on visuals it never describes. Each of these is small, and each one is the kind of detail audiences and clients notice.
7:48 Recap
Recap. Captions are essential for sound-off viewing, accessibility and non-native viewers, and they must be corrected, styled and provided per language, ideally as files too. Accessibility also means describing meaningful visuals, avoiding flashing, and plain language. Use a structured QC checklist covering audio, visuals, content and compliance, done by someone other than the maker, and always finish with the phone test, sound on and sound off.
8:17 Try this now
Try this now. Take one of your recent videos. Run the full QC checklist from the lesson, including the phone test with sound off. If you have an SRT file, run the caption check script on it. Fix at least two issues you find, and note which checklist line caught each one. That note tells you which lines of the checklist are earning their place.
Accessibility is quality
Captions are not optional extras. Many people watch video with the sound off, and viewers who are deaf or hard of hearing rely on captions entirely. Good captions also help non-native speakers and can support discoverability. Accessibility laws and public-sector requirements in several markets reinforce this, and it is simply good practice.
Captioning workflow
- Auto-generate captions using your editor, avatar platform or a transcription tool.
- Correct them. Automatic captions often err on names, brand terms, numbers, local words and mixed languages such as Urdu-English.
- Style for readability: - High contrast (light text with dark outline or background, or the reverse). - Large enough for mobile. - One or two lines at a time, broken at natural phrases. - Positioned away from platform UI (bottom buttons, side icons) and away from faces.
- Timing: captions appear with speech and stay long enough to read.
- Language versions: caption each dubbed version in its own language; do not leave English captions on an Arabic dub.
- Upload caption files where platforms support them (for example on YouTube) as well as, or instead of, burned-in captions, so viewers can toggle them and platforms can index them.
Other accessibility practices
- Describe key visuals in narration when they carry meaning ("As you can see on the chart, sales doubled" becomes "Sales doubled from March to June").
- Avoid rapid flashing effects that can harm people with photosensitive conditions.
- Color contrast for on-screen text.
- Plain language: short sentences help everyone, including viewers using translation tools.
- Alt text for thumbnails and images in posts where platforms support it.
Quality control checklist for AI media
Run this for every video before publishing.
Audio
Visuals
Content
Compliance
Final
Why "watch on a phone with sound off"?
Most short-form viewers watch on phones and often without sound. This final pass reveals captions hidden under platform buttons, text too small to read, and whether the video still makes sense silently.
Worked example
A Dubai fitness studio's AI-dubbed Arabic promo passes the audio check, but the sound-off phone test shows captions still in English and positioned under the like and comment buttons. They regenerate Arabic captions, correct two misspelt class names, move captions up, and publish. That ten-minute check prevented a confusing launch.
Hands-on: build and check an SRT file
If your platform accepts caption files, upload an SRT rather than relying only on burned-in text. A minimal SRT looks like this:
1
00:00:00,000 --> 00:00:02,600
Your Reels get views,
but no DMs?
2
00:00:02,600 --> 00:00:05,200
Here's why, and the fix
takes one line.A quick automated check catches the most common problems before a human review:
import re
MAX_CHARS, MAX_LINES, MIN_SEC = 42, 2, 1.0
ts = re.compile(r"(\d+):(\d+):(\d+),(\d+) --> (\d+):(\d+):(\d+),(\d+)")
def secs(h, m, s, ms):
return int(h) * 3600 + int(m) * 60 + int(s) + int(ms) / 1000
blocks = open("captions_en.srt", encoding="utf-8").read().strip().split("\n\n")
for block in blocks:
lines = block.splitlines()
t = ts.match(lines[1])
start, end = secs(*t.groups()[:4]), secs(*t.groups()[4:])
text = lines[2:]
if len(text) > MAX_LINES:
print("too many lines:", lines[0])
if any(len(line) > MAX_CHARS for line in text):
print("line too long:", lines[0])
if end - start < MIN_SEC:
print("too fast to read:", lines[0])It does not check spelling of names; a person still has to read every caption.
QC roles and timing
Assign QC to someone who did not make the video, and time-box it: roughly the video's length twice (one uninterrupted watch, one checklist pass) plus fixes. For multilingual releases, each language needs its own reviewer.
Second worked example: an e-learning provider in the UK
A UK training company publishing to public-sector clients adopts a caption standard (sentence case, two lines, speaker labels where there are two voices, sound cues such as "[music fades]" where meaningful) and adds an accessibility statement to each course page. Their QC checklist gains one line: "Would this make sense to someone who cannot hear it and someone who cannot see it?" That question catches narration that says "as you can see here" without describing what is shown.
Pitfalls
- Trusting auto-captions without reading them.
- Burned-in captions only, with no caption file, where the platform supports one.
- QC by the same person who made the video, when a fresh pair of eyes is available.
Key takeaways
- Captions are essential for sound-off viewing, accessibility and non-native speakers.
- Always correct auto-captions, especially names, numbers and mixed-language speech.
- Use a structured QC checklist covering audio, visuals, content and compliance.
- Do a final phone check with sound on and off before publishing.
Check your understanding
Quick questions to lock in the lesson. They don’t count towards your certificate.
Put it into practice
Run the full QC checklist on one of your recent videos and fix at least two issues it reveals.
Enrol for free to save your progress
Reading is always free. Enrol to keep your place, take the final assessment and earn a verifiable certificate.