⭐ Storyboard MCP · Flagship · Video translation
Camille, en trois languesOne French speaker, three target languages — same voice, same tone, same pacing. One MCP call per language. ~$4.50 end to end.
A short personal story in French, translated to English, Spanish, and Chinese via heygen-translate-speed. The speaker's own voice is cloned and re-rendered in each target language; lip-sync is regenerated from scratch. No talking-head synthesis, no transcript editing, no After Effects. Paste a video URL, name the targets, ship the localized cuts.
Source — French, the original
🇫🇷 Source · FrenchCamille's grandmother's sea stories — Pacific coast
Three translations — same voice, same pacing, three languages
Play any two side-by-side. The voice you hear in English is the same voice from the French source — same warmth, same hesitations, same Mediterranean musicality. HeyGen v2 didn't synthesize a generic narrator; it cloned Camille from her own 30-second clip and made her bilingual on the spot.
Pipeline trace — one model, one call per language
Stage 1 — SOURCE
Any publicly-fetchable video URL. For this showcase: a 30s static-portrait MP4 hosted at
raw.githubusercontent.com/livepeer/storyboard/.../camille-source-fr.mp4
(Production: yt-dlp <youtube_url> → push to /public → use the raw URL.)
Stage 2 — TRANSLATE (parallel × 3)
For each target: create_media({action:"animate",
model_override:"heygen-translate-speed",
params:{video_url:<source>,
output_language:"English" | "Spanish" | "Chinese"})
→ 3 MCP jobs, run concurrently → all 3 done in ~3-4 min wall-clock
Stage 3 — DOWNLOAD
Each job returns a v3b.fal.media/.../output.mp4 URL.
curl -o public/chapters/multilingual-demo/camille-<lang>.mp4 <url>
Output → 3 MP4s, 1080×1080 each, ~1.3 MB · same speaker, three languages
QC pass — what to look for in a video translation
| Target | Voice match | Lip-sync | Register | Proper nouns | Duration drift |
|---|
| English | ✓ Same voice | ✓ Clean | Conversational, warm | "Camille" preserved | +1.2s acceptable |
| Spanish | ✓ Same voice | ✓ Clean | Conversational, warm | "Camille" preserved | +0.4s tight |
| Chinese | ✓ Same voice | ✓ Clean | Conversational (Mandarin) | 未本地化（保留 "Camille"） | +1.8s acceptable |
Cost — honest end-to-end
Source hosting
$0.00
raw.githubusercontent / push to /public
EN translation
$1.50
heygen-translate-speed · 30s
ES translation
$1.50
heygen-translate-speed · 30s
ZH translation
$1.50
heygen-translate-speed · 30s
Total: ~$4.50 for a 30-second story in three languages, voice-cloned, lip-synced. Compare to an agency dubbing rate ($50–$200 per minute per language). The per-second cost is $0.05/s/language.
Reproduce — the playbook is open
Pipeline lives in the video-translate-multilingual playbook. The exact command:
curl -X POST https://sdk.daydream.monster/inference \
-H "Authorization: Bearer $DAYDREAM_API_KEY" \
-H "Content-Type: application/json" \
-d '{"capability":"heygen-translate-speed",
"params":{"video_url":"<any public URL>",
"output_language":"English"}}'
⚠ output_language must be the English NAME (capitalized): "English", "Spanish",
"Chinese", "French" — NOT ISO codes "en", "es", "zh", "fr" (those return 422).