Korean 자막 files, cued and downloadable.

Korean does put spaces between phrases, but the rules are their own thing: 띄어쓰기 governs where a space belongs, and the boundary generally falls between a noun phrase with its attached particle and the next unit. That matters for captions because the spaces determine where a renderer may wrap. A caption split at the wrong space separates a noun from the particle that tells the reader its role, which is jarring even when the timing is right.

Hangul is built in syllable blocks, each of which occupies one full-width slot however many letters it contains. Line-length budgets therefore look nothing like Latin-script ones; Korean caption practice tends to sit near sixteen characters a line over two lines. Honorific verb endings stretch a sentence noticeably — the polite forms add syllables at the end of every clause — so formal speech needs more cues than casual speech for the same minute.

• Loanwords are written in hangul as they are pronounced in Korean, which is what a viewer reads, rather than in Latin script.
• Hanja rarely appears in modern speech-derived captions and is not inserted where the speaker did not use it.
• Numbers have two counting systems, native and Sino-Korean, and the cue text follows whichever the speaker used.
• Full-width punctuation and spacing both influence how many blocks fit on a line, so measure on screen rather than by character count alone.

The SRT and VTT downloads are cue text and timings, with no name attached to any line. Voices are separated in the transcript view, and the JSON export from the same job carries that separation for downstream use.