Korean burned-in captions in a CapCut style: Sume Hangul styles

For Korean speech burned into a video, Sume ships six Hangul caption styles and 29 fonts through one API call, $0.20 per clip up to 60 seconds. The default for Korean text is black-outline, which the docs describe as CapCut-style white text on a thick black outline in the middle of the frame. I could not read CapCut's own pricing page (the request returned a 404), so this post makes no claim about CapCut features or cost; it only describes what Sume does.

The main trap is the style/language mismatch. Latin styles slam, punch and tiktok-green have no Hangul glyphs, and Sume rejects Korean text sent to them with a 400 instead of rendering empty boxes.

Karaoke ad look and fonts

korean-ad is a separate ad-karaoke style: one short phrase at a time in the lower third, with the spoken word at a heavy weight. It is not what you get by default, so select it and set language to ko. Fonts apply to Hangul styles only: pretendard is the baseline, do-hyeon is the thick rounded classic, black-han-sans and gasoek-one are impact faces, and gmarket-sans is the retail display staple.

All 29 fonts use the SIL Open Font License 1.1 and are bundled in the renderer. Names outside the list are rejected, and there is no silent fallback face. Weight travel on weight-shift and korean-ad works only with Pretendard, the one font with a weight axis.

Control and correction

Add design overrides for one request: colors, typography, placement, phrasing and motion. Numbers outside their documented range return a 400 at request time, so a bad look fails before you pay. Send script_text if you want the exact spelling of product names: Sume keeps the speech-recognition timings and aligns your text to them, and typed errors such as script_alignment_mismatch tell you when it cannot.

To try another style, send source_caption_id with a new style. Sume reuses the stored word timings, so speech recognition does not run again, and the $0.20 price is unchanged.

• Send language: "ko" as a hint to speech-to-text; it never selects the style or the font.
• Silent clips fail with caption_no_speech; send cues with text, start and end instead.
• Video must be a public HTTPS URL; private or signed URLs are rejected.

Verdict

If your captions are made in a phone editor by one person, use that editor. If you need the same Korean look on a hundred ads, produced from code with exact spelling control, the six styles and the restyle path are the reason to call an API.