{
 "task_id": "20260914_0227_姿态参考测试2_三人不重复",
 "requirement": "按参考帧姿态复刻：主角居中抬手，左一女弯腰、右一女弯腰，三个不同配角各出现一次",
 "params": {
  "width": 1280,
  "height": 720,
  "duration": 5,
  "seed": 20260917,
  "steps": 8,
  "fps": 24
 },
 "vision_description": "（外部传入提示词，未识图）",
 "prompt": "subject_definitions:\n<Subject 1> is the young Chinese woman in <Picture 1> (front view): fair skin, dark brown hair in a high ponytail with loose strands, wearing a white short-sleeve school polo shirt with blue collar and blue side panels, and dark navy track pants with white side stripes. <Picture 1> is the front-view identity anchor for <Subject 1>.\n<Picture 2> is a character turnaround sheet showing the SAME single woman from three angles - the three figures are ONE person, not three different people. Costume details only.\n<Subject 2> is the young woman in <Picture 3> (front view): twin dark ponytails, black cat-ear hair accessories, a white short-sleeve T-shirt with black trim, black shorts, white over-the-knee stockings. <Picture 3> is her identity anchor.\n<Subject 3> is the young woman in <Picture 4> (front view): long straight black hair, a long white high-collared ao dai style dress with long sleeves over white trousers. <Picture 4> is her identity anchor.\n<Subject 4> is the environment in <Picture 5>: a vast open white-jade viewing terrace above a sea of clouds, an ancient pine reaching in from the left, a bronze incense burner on a carved balustrade at the right, distant palace roofs in mist, cool blue-grey sky with warm golden sunset light from the right. <Picture 5> is the scene anchor.\n<Picture 6> is a POSE, STAGING AND CAMERA REFERENCE ONLY: it defines the camera height, shot size, where each person stands and the exact body poses. The featureless 3D mannequins inside <Picture 6> must NOT appear and must never be copied; replace them with <Subject 1>, <Subject 2> and <Subject 3>, keeping their own faces, hair and costumes.\nAll human subjects are photorealistic real people - not 3D mannequins, not anime, not illustrated.\n\nsummary:\n[reference generation] A 5-second single continuous take with no cut, reproducing the framing, staging and body poses of <Picture 6> with exactly THREE women on the terrace of <Subject 4>. <Subject 1> stands centre facing the camera with the right arm raised and the elbow bent, hand at shoulder height, the left arm relaxed. <Subject 2> stands to the left of frame and <Subject 3> stands to the right of frame, both bent forward at the waist with heads lowered and arms hanging down, as in <Picture 6>. The camera is static at chest height. The cloud sea drifts behind the balustrade. Nothing enters or leaves the frame.\n\nretention_analysis:\n<Subject 1> (appears in [Shot 1]): fully_preserved - face, ponytail, white-and-blue school polo, navy track pants; raised-arm pose follows <Picture 6>.\n<Picture 1> ([Shot 1] identity): fully_preserved.\n<Picture 2> ([Shot 1] costume reference): partially_preserved - costume details only; the three views are one character.\n<Subject 2> (appears in [Shot 1]): fully_preserved - twin ponytails, cat-ear accessories, white T-shirt, black shorts, white stockings; appears EXACTLY ONCE, on the left.\n<Subject 3> (appears in [Shot 1]): fully_preserved - long straight black hair and the long white ao dai over white trousers; appears EXACTLY ONCE, on the right.\n<Subject 4> (appears in [Shot 1]): fully_preserved - terrace, pine, balustrade, incense burner, cloud sea.\n<Picture 6> (appears in [Shot 1]): pose_reference_only - framing, staging and body poses reproduced; the mannequin figures are discarded and NOT rendered.\n<Picture 6> mannequins and overlays: not_referenced - the red and white mannequins, and any text, logo or watermark, must never be reproduced.\n\ndetailed_description:\n[Shot 1] Photorealistic cinematic footage. Horizontal 16:9. One unbroken take for the full 5 seconds.\n00:00-00:05.000: static shot at chest height matching the framing and staging of <Picture 6>. Exactly three women are present, each appearing once. <Subject 1> stands at the centre of the jade terrace of <Subject 4>, framed from head to mid-thigh, in the same pose as the central figure of <Picture 6>: torso facing camera, right arm raised with the elbow bent and the hand near shoulder height, left arm relaxed. She holds the raised hand and moves subtly with the beat - the shoulders sway, the raised hand rotates a little, the head tilts, the ponytail swings. <Subject 2> stands at frame left and <Subject 3> at frame right, both in the bent-forward posture of the outer figures of <Picture 6>: waist bent, head lowered, both arms hanging down; they hold the pose and sway gently. There are exactly three women in total - do not add, duplicate or quadruple any of them; each costume appears only once.\nThe featureless 3D mannequins of <Picture 6> must never appear; only their poses, staging and camera framing are reproduced. No text, no subtitles, no watermark, no logo, no UI overlay anywhere. No 3D mannequin, no anime, no illustrated look, no extra person, no duplicated person, no camera shake, no green screen.\n\noverall_soundscape:\nHigh-altitude wind, faint cloth fluttering, soft footsteps on stone. No dialogue.\n\nnon_diegetic_music:\nA steady mid-tempo electronic dance beat with a soft kick on every beat and a light synth pulse, consistent from start to finish.",
 "video_url": "http://192.168.31.76:8899/10-视频生成流水线/20260914_0227_姿态参考测试2_三人不重复/20260914_0227_姿态参考测试2_三人不重复_v1.mp4",
 "time": "2026-09-14 02:33:51"
}