Korean restaurant reel captions in Gmarket Sans: one font field, $0.20

To burn Korean captions on a restaurant reel in Gmarket Sans, send the clip to Sume's video caption job with style black-outline, font gmarket-sans and language ko. The job costs $0.20 for videos up to 60 seconds. Gmarket Sans is described in the docs as geometric retail display, the CapCut and YouTube title staple, which suits menu and price text.

How the three fields interact

The docs are explicit: style selects the look and the motion, font selects the face, and language only tells speech-to-text which language to expect. Language never selects the style or the font. If you omit style, Korean text resolves to black-outline, not to the Latin default that would burn Korean as empty boxes.

Hangul fonts only work with Hangul styles. Sending a Hangul font with a Latin style such as slam returns 400 caption_font_requires_hangul_style, and sending Korean text to slam, punch or tiktok-green returns 400 caption_hangul_text_latin_style. These errors cost nothing; the render never starts.

A request

Send the clip as a public HTTPS video_url and an Idempotency-Key. Use script_text if the owner reads a fixed line such as a daily special, so the burned words match the menu spelling instead of the transcript.

If the clip has no speech, such as a plating shot with music, the job fails with caption_no_speech. Send cues with text, start and end to burn authored overlay text without speech-to-text.

curl -X POST https://api.sume.com/v1/video-captions \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: reel-gmarket-001" \
  -d '{
    "video_url": "https://media.sume.com/artifacts/example/reel.mp4",
    "style": "black-outline",
    "font": "gmarket-sans",
    "language": "ko"
  }'

Trying another face

Restyling does not transcribe again: send source_caption_id with a new font and Sume reuses the stored word timings. The price does not change because a restyle is still a render, so try two faces and keep the one that reads better at phone size.

The fonts are open-licensed and bundled in the renderer, and the API rejects any name outside the list rather than falling back to another face.

A practical note

Check the reel at phone size before publishing. Small Hangul with a thick outline can clog at 12 pixels, so view the burned file on a phone, and if it is hard to read, try the clip-wipe style, which the docs call the clearest at small phone sizes, or raise typography.font_size_ratio in design. A restyle with source_caption_id reuses the word timings, so each trial renders without a second transcription. Keep menu names in script_text spelled exactly as printed on the menu, so the burned text never differs from the board.

Before you scale

Every price in this post is a Sume list price read from the public catalog on 2026-10-09, and the arithmetic is shown so you can redo it with your own counts. Rates can change, so re-read the catalog before a large batch and run a small test first. Sume bills the job's captured amount, and the job result tells you what was used, so compare the first run's receipt with your estimate before you scale the work to the full set.