PROMPT RECORD图像记录

给 Claude Code 一条这样的简短指令（如图所示），它就会返回一段超级详细的时间戳提示词，像这样。然后你把这段提示词连同你的视频一起交给 Seedance 2.5。就这些： REFER...

中文说明

给 Claude Code 一条这样的简短指令（如图所示），它就会返回一段超级详细的时间戳提示词，像这样。然后你把这段提示词连同你的视频一起交给 Seedance 2.5。就这些： REFERENCE INPUTS <<>> = 参考表演视频。 <<>> = 一张用于参考机位和构图的无人机取景静帧，仅用于机位和前景枝叶构图：俯视的远摄长焦视角，从大量深色、严重失焦的树叶和树枝之间拍摄下方的桌子旁的独自一人。仅从该图取其机位和这种前景枝叶的取景方式。其地点、人物、道具、色彩与光线不得出现在输出画面中。 下方的机位顺序与时间戳对应所提供 13.334 秒素材片段的前 7.000 秒，分辨率 1920x1080，30 fps。 MASTER DIRECTION 创建一段...

原始 Prompt

Give to Claude Code a short instruction like that (Image attached) and It will return a super detailed timestamps prompt like that. And you give this one to Seedance 2.5 with your video. That's all: 

REFERENCE INPUTS
<<<video>>> = the reference performance video.
<<>> = a drone-framing reference still, used for camera vantage and composition
only: an elevated, long-lens view looking down at a lone man at his table, shot through a
mass of dark, heavily out-of-focus leaves and branches that frame the edges of the picture.
Take nothing from it but that camera position and that foreground-foliage framing. Its
location, its people, its props, its colour and its light must not appear in the output.
The camera order and timestamps below follow the first 7.000 seconds of the supplied
13.334-second source clip, 1920x1080, 30 fps.

MASTER DIRECTION
Create one photorealistic, cinematic 7.000-second, 16:9 video containing the four shots
below, separated by three clean hard cuts at the specified times. Recreate the same garden
piece-to-camera as <<<video>>>, as if all four camera setups captured the same take
simultaneously. The performance continues through every cut: output time t depicts the
corresponding reference time t. Preserve the original speech, speed, gestures, pauses,
gaze changes and action order, and keep the original voice and location audio in sync. Do
not restart the performance when the camera changes. Keep every cut time exactly; end at
7.000 seconds without speeding up, freezing or adding a fade.

CONTINUITY
Use <<<video>>> for the single man on screen: his exact face, short dark cropped curly
hair, full dark beard and mustache, age and build; his olive-green quarter-zip knit sweater
with the small embroidered chest logo, the black t-shirt collar beneath it, the small black
clip-on wireless microphone at that collar, and his light-blue jeans. Preserve the open
dark grey laptop resting on his thighs, the light wooden garden bench with its thick cream
cushions and matching cream-topped wooden side seat at frame right, the horizontal wooden
slat fence, the tall bamboo at the left, the dense variegated broadleaf shrubs behind and
right, the small potted plant beside the seat, and the soft, cool, overcast daylight with
no harsh shadows.
The camera may change position, height and screen direction; the man, the bench, the laptop
and the planting retain their fixed physical positions in the garden. Match object states
across every cut: laptop lid angle, hand positions and contacts, the microphone, the
sweater's folds, and where his gaze is pointed at that instant. Do not add another person,
a new prop, a second laptop, or any element absent from <<<video>>>.

TIMECODED CAMERA SEQUENCE

01 | 0.000-1.302 seconds | Original frontal medium, locked off
Begin exactly on the reference framing: a static eye-level medium shot, the man centred,
seated on the bench with the open laptop on his lap, the bench arm and cushions filling the
lower left, the cream side seat at frame right, bamboo and shrubs filling the background.
He looks straight into the lens and begins the sentence, hands resting above the keyboard.
No camera movement, matching the reference's tripod stillness and lens. This shot exists
only to establish the original angle before it is replaced.

02 | 1.302-3.903 seconds | Elevated drone view, through the foliage
Hard cut on the exact frame he says the word "any", so the new angle is already fully in
place on that word. Move to a high drone-style camera hanging well above and in front of
him, looking down at roughly 45-55 degrees, positioned as in <<>>: a near
foreground mass of dark, heavily out-of-focus leaves and branches crowds the top, the left
edge and the lower corners, so the shot reads as though filmed from inside a tree canopy.
He sits small in the clean centre gap, alone, still in the olive-green quarter-zip with the
laptop open on his lap, and from this height the whole set is revealed around him: the light
wooden bench, the cream cushions, the cream-topped side seat, the slat fence, the paving and
the beds of bamboo and broadleaf shrubs. Long-lens compression, a slow gentle drone drift
with a slight descent, no orbit and no snap zoom. He keeps speaking and gesturing toward the
original camera position, which sits below and in front of him, so he never looks up into
this camera. Keep the garden's own cool, soft, overcast daylight and its natural colour: do
not borrow the warm hazy grade of <<>>.

03 | 3.903-6.055 seconds | Close shot
Hard cut to a tight eye-level close-up: head and the top of his shoulders only, shallow
depth of field, the greenery behind him dissolved into soft blur. Cover the pause and the
spoken line about distance, and hold through the short silence that follows it. Keep his
facial proportions, beard and hairline exactly as the reference, with his gaze returning to
the lens. Nearly fixed framing, no zoom.

04 | 6.055-7.000 seconds | Extreme wide, elevated
Hard cut on the exact frame he begins the words "any scale", so the change of scale lands on
the word. Move to a high, distant wide shot of the whole garden from above and behind the
original camera: the man and the bench occupy a small portion of the lower-centre frame,
with the full fence line, the bamboo, the paving and the surrounding planting revealed
around them. He remains mid-sentence and mid-gesture at natural scale for this distance.
Hold a stable, gently floating composition; no crash zoom. Hold this framing through the
short silence after his line and end cleanly on it at 7.000 seconds, with no fade, no push
in and no speed change.

FINISH
Preserve the reference's natural photographic texture, cool daylight colour, light
direction, lens character, motion blur, focus behaviour and original audio through 7.000
seconds. Keep the drone-canopy framing only in shot 02. No additional cuts, transition
morphs, invented actions, altered wardrobe, duplicated limbs or props, added titles,
subtitles, captions, graphics, music, slow motion or speed changes. One finished
7.000-second video; one performance reference and one framing reference only.