The Master Guide / 10 Voice and lip sync · Speaking shots and lip sync
Route 1, for close-up hero lines
Seedance hears the approved recording and acts to it; the exact take still ships, through Sync-3, inside a timing gate.
Seedance hears the approved recording and acts to it; the exact take still ships, through Sync-3, inside a timing gate.
Route 1. Option: use it when a close-up hero line needs body and breath to ride the voice (a laugh breaking into the line, a sob, a shout, a punchline whose timing is the point) [single strong source]. Each half of the route has support; the two together have not yet been shown. Name each shot that uses it in the plan and price it there; make it only when you choose it for that line. Otherwise the shot stays on the default.
How it runs
Settings: the take as an audio reference, audio on, the anchors as image references, the line quoted in Arabic letters.
The reference is voice only and at least 2 s long; a shorter line uses a copy padded with room tone.
The timing, outside the quotes, from the measured take: the speech length to 0.1 s, never the padded length, and its window ("the speech in the audio lasts 2.1 s and fills the clip from 1.0 to 3.1 s").
Mirror the take's pauses with an ellipsis inside the quote, for example:
«لأ… مش دلوقتي.»
The duration covers the speech window plus a tail: at least 4 s.
The face stays within about 45° of the lens during the line, because here Seedance performs it.
The prompt is Template E.
Discard the returned audio. Run Sync-3 with a plate-length file, the take placed at the clip's measured onset (Seedance re-times the line), and cut_off.
The timing gate (working numbers): the overall span within ±5% of the take's; every pause inside the line and the emphatic beat within ±3 frames, from the word timestamps of both. Measure on the draft and again on the finish. A take outside the gate sends the shot back to the default plate.
Draft Mode with audio references is unconfirmed. If it refuses, draft at 720p (7 Higgsfield credits a second; prices and their dates in Speaking shots and lip sync → Attempts and budget per line).
A refusal with an audio reference: use a clean, voice-only reference.
The Franco quote
A named variant of Route 1 [single strong source], used only where you choose it for an Egyptian line:
It is the audio-driven plate with one change: the quoted line is the Franco line (Arabic written in Latin letters, sometimes called Arabizi) from the project's spelling sheet.
Egyptian only. Other dialects are not yet tried, and a native reviewer would write their sheet.
It rests on one practitioner's reel, and what that reel ships hasn't been confirmed: listen to it with sound before you rely on it.
The exact take through Sync-3, inside the gate, is the route that ships (its outcome (b)). Seedance's returned Egyptian voice is its outcome (a); like any returned voice on an Arabic line, it is discarded and never shipped (decided; see Speaking shots and lip sync → The re-voiced read).
Drop it for that shot on misreads, or an English accent that one respelling round doesn't fix; on lips after Sync worse than the plain audio-driven plate's; or on refusals.
The spelling sheet
You write it from the approved take; a native ear corrects any word. It never goes to a voice engine.
# | Rule | Example |
1 | Latin letters, apostrophes and ordinary punctuation; straight double quotes around the line | — |
2 | No digits at all: none of the 2, 3, 5, 6, 7, 8 or 9 of chat Arabic | — |
3 | The ain is left out. If the returned speech skips a needed beat, try a hyphen there and note it | "ayezha", "Araby" |
4 | The glottal stop is written as an apostrophe: the Egyptian qaf, and a hamza inside a word | "teb'a" |
5 | Short vowels a, e, o; a word-final vowel single; a long vowel inside a word as aa, ee, oo; doubled consonants doubled | "teb'a", "feen", "haat", "huwwa" |
6 | Jeem as g; tha as t or s and dhal as z or d, as the word is said; the deep ha as h; kha as kh; sheen as sh; ghain as gh (if it drops, respell that word and note what works); the emphatic sad, dad, ta and za as s, d, t, z | — |
7 | Watch English letter pairs (zh, th, ou or ow, ch); split a misread one with a hyphen | "ayez-ha" |
8 | Respell any word that reads as an English word | "feen", not "fine"; "haat", not "hate" |
9 | English brand names keep their English spelling | — |
10 | One fixed spelling per word for the whole project, kept in the pronunciation sheet | — |
Sources
Higgsfield's Lipsync Studio post (the voice drives the face); Lullaby by Keigo Matsumaru (project) (a voice bound to an image and an audio reference); abdullrh_man (reel) (the Franco quote).