参考图：角色三视图（正面 / 侧面 / 背面，同一个人）——黑长直双马尾 + 粉黄花饰与浅色发带、 薄荷绿+白汉元素连衣裙（金色描边、青色腰带飘带）、白色分离袖套/手套、白裤袜、黑色玛丽珍鞋； 背景是纯灰摄影棚（这个背景必须被替换掉）。 需求：城市场景 · 主角半身出镜 · 手持手枪指着镜头 · 表情伤心而不忍心。 用法：竖屏（9:16），时长 5s。竖屏不做裁剪，改为在提示词里用文字声明"三视图是同一人"。
节点设置：模式 = 参考生视频 / reference，比例 9:16，参考图尺寸 1k，引用方式 按序号，时长 5s。
subject_definitions:
<Subject 1> is the young woman in <Picture 1>: long black hair gathered into two low twin tails with pink-and-yellow flower ornaments and long pale ribbons at each side, fair skin, and a mint-green and white hanfu-style dress with gold trim at the waist, a teal sash ribbon hanging down the front, detached white sleeves and gloves, white tights and black Mary-Jane shoes; her face is the one shown in the front view.
<Picture 1> is a character turnaround sheet showing the SAME single person from three angles (front, side, back) — the three figures are ONE character, not three different people. Use the front view as the identity anchor for her face, and the side and back views only for the costume, hair ornaments, sleeve volume and sash details.
<Subject 2> is the target environment: a modern city street at night after rain, with wet asphalt reflecting cyan and magenta neon signs, blurred headlights and traffic lights in the background, light rain in the air and steam rising from a vent.
summary:
[reference generation] The target video is a single continuous medium close-up of <Subject 1> standing in <Subject 2>. She faces the camera, raises a compact black pistol toward the lens, and holds it pointed at the viewer with a heartbroken, reluctant expression while the camera slowly pushes in. Her identity from the front view of <Picture 1> is fully preserved; the plain studio background of the reference sheet is completely replaced by the city street, and only one person appears in the video.
retention_analysis:
<Subject 1> (appears in [Shot 1]): fully_preserved - identity_fully_preserved; the black twin tails, the pink-and-yellow flower ornaments, the long pale ribbons, the mint-green and white dress with gold trim, the teal sash, the white gloves and the white tights are all retained.
<Picture 1> ([Shot 1] identity and costume reference): partially_preserved - the front view supplies her face and costume; the side and back views only supply costume and accessory details. The three views are the same character and must never appear as three people or as a side-by-side sheet.
<Subject 2> (appears in [Shot 1]): fully_preserved - the night city street, the wet reflective asphalt, the neon signs, the blurred traffic lights and the light rain are retained.
detailed_description:
The target video is in a cinematic neo-noir style with cool cyan and magenta neon lighting, wet reflective surfaces, heavy bokeh in the background and a shallow depth of field.
[Shot 1] The shot is a medium close-up of <Subject 1> from the chest up, framed in a vertical composition. She is the young woman from the front view of <Picture 1>, with black hair in two low twin tails, pink-and-yellow flower ornaments and long pale ribbons at each side, wearing the mint-green and white hanfu-style dress with gold trim, the teal sash ribbon down the front, the white detached sleeves and gloves. She stands on a rain-wet city street at night: cyan and magenta neon signs smear across the wet asphalt behind her, headlights and traffic lights dissolve into round bokeh, and light rain falls through the frame. She faces the camera directly and extends her right arm toward the lens, holding a compact black pistol with the muzzle pointing straight at the viewer; the weapon and her forearm are slightly foreshortened as they come toward the camera. Her expression is heartbroken rather than angry: her eyes are wet and glistening, her brows are drawn slightly upward in the middle, her lips are pressed together and tremble a little, and she never smiles. The camera pushes in with small amplitude at slow speed while a subtle handheld micro-shake keeps the frame alive; focus racks from the pistol muzzle to her eyes. Her extended arm shakes very slightly, and a single strand of hair and the pale ribbons drift with the breeze. No second person and no group appear anywhere in the frame, and the plain grey studio background of the reference sheet never appears. The shot ends on her face, still holding the pistol toward the camera with the same reluctant, grieving look.
overall_soundscape: Fine rain patters on the wet asphalt and on metal surfaces nearby. Distant traffic hisses past with occasional far-off horns, and a vent releases a soft continuous hiss of steam. Her slow, uneven breathing is clearly audible close to the microphone, together with a faint rustle of fabric as her arm trembles.
non_diegetic_music: A single sustained low string note holds under sparse, slow piano notes. A soft pulse like a slow heartbeat enters as the camera begins to push in and continues to the final frame without resolving.她不开枪，全程只是指着镜头——这样既符合"不忍心"的情绪，也避免出现血腥画面。要"她最后放下枪"的版本，把最后一句换成：
At the end of the shot her arm lowers a few centimetres and the muzzle drifts off the lens as her eyes close.
第一行固定指令必须放最前面；比例记得设 9:16。
For the target video, at 0.00 seconds into the target video, <Picture 1> (from [Shot 1]) is fully referenced.
integrated_multimodal_description: [Shot 1] Live-action, cinematic, neo-noir style. The young woman shown in the reference image — long black hair in two low twin tails with pink-and-yellow flower ornaments and long pale ribbons, a mint-green and white hanfu-style dress with gold trim, a teal sash ribbon down the front, white detached sleeves and gloves, white tights — appears alone, from the chest up, in a vertical medium close-up. Note that the reference is a character sheet showing the same single person from three angles: only her face and costume are taken from it, and only ONE person appears in the video. The plain grey studio background is completely replaced: she now stands on a rain-wet city street at night, where cyan and magenta neon signs reflect on the asphalt and headlights and traffic lights blur into round bokeh in the background while light rain falls. She faces the camera and extends her right arm toward the lens, holding a compact black pistol pointed directly at the viewer, the weapon slightly foreshortened. Her expression is heartbroken, not angry: her eyes are wet and glistening, her brows drawn slightly up in the middle, her lips pressed and trembling, and she does not smile. The camera pushes in with small amplitude at slow speed with a subtle handheld micro-shake, focus racking from the pistol muzzle to her eyes; her arm trembles very slightly and the pale ribbons drift. No second person, no group and no side-by-side three-view layout appear. The shot ends on her face as she keeps the pistol pointed at the camera with the same reluctant, grieving look.
overall_soundscape: Fine rain patters on wet asphalt and nearby metal. Distant traffic hisses past with occasional far-off horns, and a vent releases a soft continuous hiss of steam. Her slow, uneven breathing is clearly audible, with a faint rustle of fabric as her arm trembles.
non_diegetic_music: A single sustained low string note holds under sparse, slow piano notes, with a soft pulse like a slow heartbeat that continues to the final frame without resolving.| 要控制的东西 | 提示词里对应的句子 |
|---|---|
| 三视图不被当成三个人 | the three figures are ONE character, not three different people ＋ No second person, no group and no side-by-side three-view layout appear |
| 灰摄影棚背景不跟过来 | the plain grey studio background of the reference sheet never appears ＋ 显式写出城市夜街的全部特征 |
| 半身构图 | a medium close-up of <Subject 1> from the chest up, framed in a vertical composition |
| 枪指着镜头 | extends her right arm toward the lens, holding a compact black pistol with the muzzle pointing straight at the viewer（+ slightly foreshortened） |
| 伤心而不忍心（不是愤怒） | heartbroken rather than angry: eyes wet and glistening, brows drawn slightly upward in the middle, lips pressed and trembling, never smiles |
| 不要出现三个人/并排画面 | 见上表第一行；另外把比例设成 9:16（竖屏本身就抑制并排） |
<Picture 1>（~/Downloads/dasiwa-h3-research/turnaround_prep.py），完整三视图降为 <Picture 2>，用双参考版提示词。1k 提到 original，或缩短时长到 4s（重建幅度更小）。retention_analysis 里把服装逐项再抄一遍，并在 detailed_description 加一句 Her costume stays exactly as in <Picture 1> and is never replaced by modern street clothing.