TikTok ad captions are white and fixed; burned-in captions differ

Two different things are called captions here, and TikTok's page only restricts one of them. For non-Spark ads, TikTok says ad captions are displayed in white with a uniform font that cannot be customized (TikTok page, read 2026-10-09). That is the text field that sits with the ad. Sume's video captions API burns text into the pixels of the video itself, with styles such as slam, punch and tiktok-green. The page does not say anything about text inside the video file, so do not assume it forbids it, and do not assume TikTok approves it either.

What the TikTok page says

For Spark Ads, the page says captions are extracted directly from the organic video's captions and a maximum of four lines can be displayed. For non-Spark ads the caption is white, with a uniform font, and it cannot be customized. The page is silent on burned-in text, so it gives you no rule to quote either way.

What Sume's captions do

A caption job takes a public HTTPS video URL, a style, optional font, language, and either speech-to-text or your own text. You can send script_text, words, or cues/segments with text, start and end in seconds. The API returns a captioned video. Each accepted standalone job reserves and captures $0.20 for videos of up to 60 seconds, per the docs.

Because the burned text is in the frames, the ad caption field on TikTok does not restyle it. If you want white text with a thick black outline, black-outline is the documented safe default for Korean, and slam is the default for Latin text. The docs say that if you do not send a style, the text decides: Latin text gets slam, Korean gets black-outline.

A sensible workflow

Keep the story in the video and the offer in the ad caption. Burn the hook line or the spoken words into the video so it reads with sound off, and put the call to action in the ad caption field where TikTok controls the look. That way the part you must style is yours and the part TikTok must style is theirs.

Check the technical floors at the same time. The non-Spark page lists 9:16 at 540x960 or larger, a bitrate of at least 516 kbps, a 500 MB limit and a 10-minute limit. See the 516 kbps floor post for those checks. For Spark Ads, which pull an organic post, the Spark post covers the four-line rule.

Caveats

Captions fail with caption_no_speech on a silent clip if you rely on speech-to-text; send cues instead. If the language is Korean, use a Hangul style and not slam, punch or tiktok-green, which the docs say return a 400 for Hangul text. I did not find a TikTok rule on burned-in text on the page above, so check your account's ad review notes if a creative is rejected.