Klip Kanvas Logo
Klip Kanvas
Create
FAQVideo GenerationTikTokAdvanced

Captions and Subtitles FAQ: Auto Captions, Styles, Safe Zones and Rules

Auto captions for AI UGC ads — styles, safe zones, editing accuracy, multilingual subtitles, accessibility, and what Meta and TikTok actually allow on screen.

Updated 2026-03-1113 min read
EnglishEspañolItalianoNorskDanskÍslenska

Most paid social is watched on mute, so captions are not decoration. This FAQ covers auto captions, styles, safe zones, accuracy and editing, multilingual subtitles, accessibility, and the platform rules that get burned-in text rejected.

01

Auto Captions

1.Does Klip Kanvas burn captions into the video automatically?

Yes. Auto captions are generated from the script and aligned to the voiceover, then burned into the Klip Kanvas export so they play even when the viewer has sound off and even when the platform’s own caption layer is disabled. That is the right default for ads: you cannot rely on TikTok or Meta to show their native captions on paid placements the way they do on organic. You can still turn burned-in captions off for a clean master, or hide them on a version you will caption in the ads manager. Generate once, then decide per placement.

#Auto Captions

2.Should I use burned-in captions or platform captions?

Burned-in for paid ads, platform-native for some organic posts. Paid placements are inconsistent about showing auto-generated captions, and you cannot A/B a hook if half the impressions never see the words. Burned-in also lets you style the first two seconds as part of the interrupt. The cost is flexibility: a typo is in the file until you re-export. A practical workflow is burned-in for 9:16 paid, and a clean export plus native captions for organic Spark-style posts where you want the app’s default look. Do not double-caption; stacked systems look like a mistake.

#Auto Captions#Platform Rules

3.How fast are captions available after a render?

They come with the render. A typical 30-second ad finishes in 3–5 minutes and the caption file is already aligned; you are not waiting on a second transcription job. If you edit the script after a render, re-export — captions will not magically rewrite themselves on the old file. First ad from sign-up still sits under 10 minutes including captions, which is why they should be on from test one. Shipping mute-unfriendly tests “just to see the avatar” wastes the only impressions that teach you whether the hook reads.

#Auto Captions

4.Can I caption only the hook and not the rest?

You can, but it is usually a bad idea. Mute viewers who stay past two seconds still need the proof beats. A better pattern is a larger, punchier style on the hook line and a quieter style on the body, so the interrupt is readable at a glance and the rest does not turn into a karaoke wall. If a legal disclaimer must appear, put it in a static lower line, not in the word-by-word caption stream. Mixing “no captions” after the hook with a talking-head is how hold rate dies on mute.

#Auto Captions#Caption Styles
02

Caption Styles

5.What caption style works best for UGC ads?

Large, high-contrast, centre-weighted, one to six words on screen at a time. The style that wins on TikTok organic — huge centre words, occasional pop emphasis — also works in paid if you keep it from covering the face. All-caps walls, thin outlines on busy B-roll, and script fonts fail on a phone. Match the brand kit for colour, but do not let brand guidelines force 12-point type. Captions in ads are a performance surface. If a designer and a media buyer disagree, test two styles on the same hook rather than arguing in Figma.

#Caption Styles

6.Should every word be on screen, or just keywords?

For ads, prefer short phrase groups, not karaoke-every-syllable and not keywords-only. Karaoke is tiring over 30–60 seconds. Keywords-only leaves mute viewers to guess the sentence. Group the line the way a person would read it: “I stopped rebuying this” as one beat, not six flashes. Emphasise one word in the hook if it carries the number or the mistake. If the spoken line is too long to caption cleanly, the line is too long to say — fix the script. Captions will not rescue a 28-word sentence.

#Caption Styles

7.Can I match captions to my brand fonts and colours?

Yes, within the caption style controls and the brand kit. Pull fill, stroke and a close font match so the words feel like the same brand as the landing page. Perfect webfont parity is a real limit — some licensed desktop fonts cannot be embedded, and we will substitute a near cut rather than pirate the file. If legal requires a specific typeface on claims, check a still frame before you scale. Colour contrast matters more than brand fidelity: pale grey on a bathroom tile background is how captions disappear.

#Caption Styles

8.Do animated captions help hook rate?

A little, when the motion is a simple pop or underline on the interrupt word. Continuous bouncing, per-letter spins and sticker storms help you look like every other template in the feed, which is a fatigue accelerant. If frequency is already heading toward 2.5–3.5, a louder caption animation is not a refresh; a new hook line is. Test static versus one modest motion on the same opening. If hook rate moves and CTR does not, you made the words more visible to people who still will not click.

#Caption Styles
03

Safe Zones

9.Where should captions sit so they are not covered by UI?

Keep them in the vertical centre-safe band: away from the top username/search chrome and away from the bottom caption, CTA and account bar. On TikTok and Reels that bottom stack eats a surprising amount of 9:16. Centre or slightly upper-centre is safer than lower-third, which is where most TV-trained editors still put type. Preview 9:16, 1:1 and 16:9 separately; a safe 16:9 lower-third is often illegal in 9:16. If you must show a price or URL, do not put it in the bottom 20% of the frame.

#Safe Zones

10.Do safe zones change between 9:16, 1:1 and 16:9?

Yes, a lot. 9:16 is the strictest because platform UI lives on the image. 1:1 is more forgiving in-feed. 16:9 for in-stream or YouTube has different lower-thirds and often more time to read. Exporting all three from one project does not auto-reflow captions perfectly for every crop — check the first frame and the CTA frame in each ratio. If a word is cropped, move the style, do not just shrink the type until it is unreadable. Re-export the ratio that failed; do not stretch a 9:16 file into 1:1 and hope.

#Safe Zones

11.Will captions cover the avatar’s face or the product?

They can, and you should treat that as a layout problem, not a surprise. Talking-head UGC wants the face clear; demo UGC wants the product clear. Shift captions above the head, or cut to B-roll when the product is on screen and keep captions shorter. A caption sitting on a product label during a demo is how you hide the one thing the ad had to show. If a scene is a close-up of texture or fit, consider hiding captions for that beat and bringing them back on the talking-head. Watch once with the sound off.

#Safe Zones

12.How do I preview safe zones before I spend ads budget?

Use the editor’s platform overlay and then a phone. Desktop preview lies: you sit closer and you ignore chrome. Full-screen the 9:16 on a device, thumb over the bottom right where the like stack lives, and read the hook. If you cannot get the first line in one glance, the type is too small or too low. Do this on the six-creative first batch, not after you scale. A typical render is 3–5 minutes; there is no production reason to skip a mute-phone pass.

#Safe Zones
04

Accuracy & Editing

13.How accurate are auto captions?

They are script-based, so they follow what you wrote rather than a noisy transcription of the audio. That is a strength for product names you spelled in the script, and a weakness if you changed pronunciation without changing the words. Numbers, SKUs and invented brand spellings still need a human glance. Slang the model “helpfully” regularises can also drift from the voice. Budget sixty seconds to read captions on every winner before it goes to ads manager. Do not QA all 40 first-drafts at the same depth; QA what you will spend on.

#Accuracy & Editing

14.Can I edit a single caption without re-rendering the whole video?

You can edit caption text and style in the Klip Kanvas project and re-export. That still costs a render, because burned-in captions are in the picture. There is no lossless “just swap the word” on a downloaded MP4 that keeps the same file hash — you replace the file. For a one-word typo on a winner, re-export is the correct path; do not paint over letters in a third-party editor unless you like quality drop. If you find yourself recaptioning every draft, the script process is too loose. Lock the words, then render the test batch.

#Accuracy & Editing

15.The captions don’t match what the avatar said — why?

Either the script was edited after generation, a pronunciation pass changed a word, or a localised voiceover is using a translation that was not written back into the caption track. Align them before you ship. Mismatch is more damaging than no captions: mute viewers think the file is spam, and sound-on viewers notice the lag. Regenerating from the current script is the fix. Do not “nudge” timing by ten frames on a 60-second file unless you like a drift that compounds. Short scenes stay in sync longer than one unbroken take.

#Accuracy & Editing

16.Should I caption filler words like um and like?

No. Leave them out of the caption track even if you left a small one in the performance for realism. Filler on screen looks like a broken file, not like a person. Same for false starts. If the voiceover is full of ums, the script or delivery is wrong — clean the read rather than captioning the mess. Captions should be the cleaned, readable version of the line, still faithful to the meaning. This is also an accessibility point: screen-reader-like reading of “um um so basically” helps nobody.

#Accuracy & Editing#Accessibility
05

Multilingual Subtitles

17.Can I generate subtitles in another language than the voice?

Yes — dual-language versions are a real workflow when you are testing a market before you recast. A Spanish caption track on an English voice can help a bilingual audience and is a poor substitute for a native voiceover with regenerated lip-sync. If you are serious about the market, localise audio, captions and claims together. Literal translation of a hook often kills it; transcreate the first line. Have a native speaker check captions even when the translation came from us. Idioms and units of measure are where auto subtitles still fail.

#Multilingual Subtitles

18.Do I need new captions when I localise the voiceover?

Yes. Generate captions from the localised script so line breaks, numbers and legal phrasing match what is said. Reusing English burned-in captions over a Hindi or Spanish read looks obviously unfinished and tanks trust. Timing will also be wrong because sentence length changes. The localisation flow keeps the avatar and edit and rebuilds voice and captions together — use that, rather than overlaying a .srt from a different language in another tool. QA the hook and the CTA in the new language on a phone before you buy traffic.

#Multilingual Subtitles

19.Which languages work best for auto subtitles?

The same 30+ languages as script and voiceover, with the same caveat: major languages are stronger than long-tail ones, and mixed-script lines (English brand names inside another language) need a look. Right-to-left scripts and languages that run much longer than English will need more manual line-breaking to stay in the safe zone. That is a real limitation, not a hidden toggle. If a launch market is one of those cases, add a native review step on the first six creatives before you batch the rest of the catalogue.

#Multilingual Subtitles

20.Can I ship the same caption file to Meta and TikTok?

You can ship the same burned-in video to both if safe zones were checked against the stricter 9:16 UI. Separate .srt sidecars are a different story — each ads manager treats uploaded caption files differently, and paid placements may ignore them. Burn-in is the reliable cross-platform path. If a platform requires a specific disclosure line, that line may need to be in the picture for that placement even if the other network hates extra text. Advertiser owns those rules; we do not certify a file for both networks at once.

#Multilingual Subtitles#Platform Rules
06

Accessibility

21.Do captions count as accessibility, or do I need more?

Burned-in captions help deaf and hard-of-hearing viewers and mute viewers, which is most of the feed. They are not a full accessibility programme. Colour contrast, readable type size, not encoding meaning in colour alone, and an accurate transcript all still matter. If you also publish on the site or YouTube, provide a proper caption file or transcript there. Do not autoplay with no captions and call it done. Accessibility is also a performance habit: the same large type that helps a hearing-impaired viewer helps a commuter with the sound off.

#Accessibility

22.What contrast and size should I use so captions are readable?

Aim for strong contrast — light fill plus a dark stroke, or the reverse — and a size you can read at arm’s length on a phone in sunlight. If you have to lean in, it is too small. Avoid placing pale type on kitchen marble or bathroom tile B-roll without a stroke or a pill background. Do not use red/green pairings to encode emphasis. Brand purple on a busy room is a common failure. When in doubt, add a semi-opaque bar behind the words for the hook only, then return to a lighter style so the rest of the ad does not look like a slideshow.

#Accessibility#Caption Styles

23.Should I include speaker labels or sound descriptions?

For single-avatar UGC ads, speaker labels add clutter and no information. Sound descriptions — [music], [cap clicks] — are useful in a transcript and noisy on a 15–30 second paid clip. Skip them in the burned-in ad unless the sound is the joke and the viewer would miss it on mute, which is rare. If you cut a longer organic piece up to three minutes, a light “[music]” at the start is enough. Keep the ad captions to speech. Extra non-speech labels eat safe-zone space you need for the claim.

#Accessibility

24.Are caption-only ads (no voice) a good idea?

Sometimes, for silent demo or text-hook tests, but they are a different format. If you already have an avatar mouthing words, run the voice; a talking face with no audio and full captions looks like a broken upload. Pure text-on-B-roll can work as a hook variant and is worth one cell in a test, not the whole account. Accessibility-wise, caption-only without audio fails users who cannot read the language quickly. Use it as a creative choice with intent, not as a way to hide a bad voiceover.

#Accessibility
07

Platform Rules

25.Will burned-in text get my ad rejected?

It can, if you pack the frame with claims, before/after text, prices that contradict the landing page, or fake native-UI chrome. Platforms also reject ads that look like they are impersonating the app’s own caption system in order to trick taps. Keep captions as speech, keep claims supportable, and do not fake a “Shop now” system button in the video. We surface common risk patterns; we cannot guarantee a given account will pass review. The advertiser owns compliance, including what the words on screen say.

#Platform Rules

26.Do Meta and TikTok have different caption rules?

The physics is the same — mute, safe zones, no fake UI — and the policy overlay is not. TikTok is stricter about looking like native organic chrome and about certain health claims in text. Meta is stricter about misleading before/after and about personal attributes in copy, which includes on-screen words. A line that is legal in voice can still fail as text, or the reverse. Check both, and do not assume a TikTok winner can be uploaded to Meta unchanged because the file already exists. Re-read the on-screen hook as if it were primary text.

#Platform Rules

27.Do I have to disclose AI in the captions?

Often you must disclose AI-generated or synthetic media, and the responsibility sits with you, not with Klip Kanvas. Meta and TikTok both require disclosure in some cases, especially when a real-looking person makes a claim, and especially in health, finance and similar verticals. Rules differ by region and still move. A caption that says the product claim but hides the disclosure in frame 47 is not a strategy. We provide toggles and tooling; we do not certify that a given caption track makes a given ad legal in a given market.

#Platform Rules

Ready to put this into practice?

Create your first AI UGC video ad in minutes — no filming, no actors, no editing.

Try Klip Kanvas free

More in this section

16 min read

Veo, Runway and Kling vs AI UGC Ad Tools FAQ

Cinematic generators like Veo, Runway and Kling are not a product-URL-to-UGC-ad factory. Compare jobs, cost per usable ad, URL workflow, and when Klip Kanvas is the ad desk.

FAQAI UGC
Read more
15 min read

Perplexity, Copilot and DeepSeek FAQ for Marketing Teams

How marketing teams should use Perplexity, Microsoft Copilot and DeepSeek for research, office work and cheaper inference — and why none of them replace a UGC ad desk.

FAQAdvanced
Read more

Ready to make ads like these?

Paste a product link and Klip Kanvas writes the script, casts the creator and renders the ad — no filming, no actors, no editing.

Try Klip Kanvas free