The inputs of a sung take
- @Image1 is the first frame.
- @Image2 is his face sheet: identity only, never shown as a shot.
- @Audio1 is the lead vocal stem cut to exactly the take's length (whole seconds, 4–15), opening on about 0.7 s of silence before his first word, everything outside the shot's own lines silenced with 20 ms fades. One or two lines per take.
- 720p (1080p for close faces), sound on, one take per job.
Template: a sung take
【Audio Policy】 The output soundtrack is @Audio1 itself: copy @Audio1 sample for sample from the first frame to the last, with the same timing, pitch and level. Do not re-sing, re-voice or re-generate it. Do not add music, background music, score, instrumental, melody, synth, pad, drone, riser, extra vocals, harmonies, ad-libs, narration or sound effects. 【Generation Goal】 @Image1 sings @Audio1: <who, where, which moment of the story; the performance in one sentence>. 【Reference Asset Roles】 Use @Image1 as the first frame. This first frame defines HAL's appearance and wardrobe, his pose, the set, the light and the camera direction. @Image2 is HAL's character sheet: use it only for his face, his platinum buzz cut, his one chrome teardrop and his headphones. It is never shown as a shot; its layout, grey background and several views are not used. He is ONE person. @Audio1 is HAL's own singing voice and the complete soundtrack of this video. His mouth follows @Audio1 exactly. 【Subject】 HAL: <the identity text>, <the look>. Only one HAL appears. 【Performance Timeline】 0-1s: @Audio1 is silent, then his voice begins; <what he does>. 1-3s: in @Audio1 he sings "<exact words>"; his lips form each word exactly when it sounds in @Audio1; <one loose action>. 3-4s: @Audio1 is silent; his mouth relaxes and closes; <one loose action>. 【Camera】 <shot size, angle, movement, start and end>. 【Core Requirements】 1. Lip-sync word by word to @Audio1; the mouth rests naturally closed while @Audio1 is silent. 2. HAL's voice is @Audio1 and nothing else. 3. Keep HAL's identity, face, buzz cut, one chrome teardrop under his left eye, headphones with the cyan light and wardrobe exactly as in @Image1 and @Image2. 4. Natural anatomy: hands with five fingers, no extra or fused fingers; normal eyes. 5. No subtitles, captions, lyrics, titles, signs, logos, watermarks or any on-screen text. 【Final Acceptance Criteria】 The soundtrack is @Audio1 unchanged; his lips match @Audio1 word by word; one HAL, on-model; no on-screen text.
Template: he doesn't sing
Same frame, with @Audio1 is the song's instrumental, used for rhythm and timing only, the tempo written out (96 BPM, a beat every 0.625 s), the movement second by second, and:
【Mouth】 He does not sing, speak or mouth any words at any moment: his lips stay closed or in a smile, or part only when he laughs.
Template: no person
A short prompt: the world, one action, the camera, "no people, no on-screen text". A board or a screen is described without numbers; the numbers are drawn in code on top.
Directing him
- One plain clause of movement per segment: "he turns to the window and back, a smile in his eyes".
- Warm by default. Mood words are obeyed literally and they stack.
- No posture rules in a take; no smile on his mouth while he sings.
- Never ask for a look past the lens or behind him.
- Nothing appears from nowhere: a hand brings it.
Sync is a path, not an offset
Seedance keeps each sung phrase at about 1× but stretches the pauses, and sometimes sings a line 10–16% slower. Align each take's own soundtrack to its window with dynamic time warping and map song time to take time along that path. A take that came back without sound is placed by his lips: shut on every b, m and p. Never show a take's first two frames.
A cheaper model that cannot hear the song
A first frame only, no sheet and no audio: keep takes to 5–8 s, write the sung words in quotes, place each take by his lips, and favour medium shots and profiles over sung close-ups.