English avatar video with Spanish and Korean subtitles via cues

Sume Avatar 1.0 speaks English, but the picture can carry other languages as subtitles. Generate the clip once, then call POST /v1/video-captions once per language with authored cues, using black-outline for Korean because the Latin styles reject Hangul.

Why cues and not speech-to-text

If you let a caption job transcribe the soundtrack, you get English. Passing cues (each with text, start, end in seconds) skips speech-to-text and burns exactly your copy, so the translated text is whatever you supply. cues, segments, words and script_text are mutually exclusive.

One job per language

Each job is billed separately at $0.20 for a clip up to 60 seconds, so two languages is $0.40 on top of the avatar clip. Use a distinct Idempotency-Key per language so a retry cannot double-bill.

Set VIDEO_URL to the public HTTPS URL of the finished avatar video (a caption job needs a fetchable URL).

import os, requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
VIDEO = os.environ["VIDEO_URL"]
JOBS = {
    "es": ("slam", [{"text": "Hola, soy tu guía.", "start": 0.0, "end": 2.2}]),
    "ko": ("black-outline", [{"text": "안녕하세요, 안내를 맡았습니다.", "start": 0.0, "end": 2.2}]),
}
for lang, (style, cues) in JOBS.items():
    r = requests.post(
        "https://api.sume.com/v1/video-captions",
        headers={**H, "Idempotency-Key": f"avatar-subs-{lang}-001"},
        json={"video_url": VIDEO, "style": style, "cues": cues},
        timeout=30,
    )
    r.raise_for_status()
    print(lang, r.json()["status_url"])

Keep the timings honest

The spoken English sets the clock. Match each translated cue to the English phrase it translates, and read the line aloud at that length; Korean and Spanish often run longer or shorter than the English they replace.