兩位動漫角色在寫實城市屋頂上，夕陽漸入藍調時刻，情感濃烈的音樂錄影帶風格影片

提示詞

這是一段 15 秒、情緒飽滿的音樂錄影帶風格影片，全程呈現於完全真人實拍、照片級寫實的屋頂場景中。

僅將 {{Image 1}} 作為角色 A 的身份參考。
僅將 {{Image 2}} 作為角色 B 的身份參考。
圖片僅供參考臉部、髮型、服裝、配件、色彩和輪廓。

只有角色 A 和角色 B 是動漫風格。
屋頂、天空、城市、燈光、道具、倒影、鏡頭光學效果和氛圍都必須看起來像真實的真人實拍影像。
不要將「アット」Image1 或「アット」Image2 的動漫風格套用到背景中。

角色規則：
僅有兩個角色：角色 A 和角色 B。
不得出現第三人、複製人、重複角色、背景人物、人形倒影或人形陰影。
不要把角色變成寫實的人類。
兩個角色都要保持造型一致，臉部、服裝、頭髮和配件穩定不變。

環境：
日落逐漸轉為藍調時刻的寫實都市屋頂。
潮濕的混凝土、直升機停機坪標線、金屬欄杆、屋頂門、管線、空調、設備箱、紅色警示燈、水窪、遠方的城市燈光，以及真實的城市天際線。
強風帶動頭髮、緞帶、袖子、裙子和飄動的布料。

嚴格避免：
不要有動漫風格的背景。
不要有插畫風格的屋頂。
不要有繪畫風格的天空。
不要有賽璐珞著色風格的建築。
不要出現無人機。
不要有文字疊加。
不要有標誌。
不要有多餘的肢體。
不要更換服裝。
不要扭曲臉部。

鏡頭 1 (0-1.5s):
屋頂全景開場。角色 A 站在左側欄杆附近，角色 B 站在右側較遠處。低角度鏡頭掠過潮濕的混凝土和直升機停機坪標線，向前滑動。

鏡頭 2 (1.5-3s):
角色 A 的中近景。她在節拍落下時轉頭看向鏡頭。頭髮和緞帶隨風飄動。背景保持照片級寫實並呈現模糊效果。

鏡頭 3 (3-4.5s):
角色 B 的中近景。她先低頭，然後抬起眼睛看向角色 A。日落的輪廓光勾勒出她的頭髮。

鏡頭 4 (4.5-6s):
雙人中遠景。角色 A 在左側，角色 B 在右側，兩人一開始就都出現在畫面中。角色 B 朝角色 A 跨出一小步。鏡頭橫向跟拍。畫面中僅有兩個角色。

鏡頭 5 (6-7.5s):
低角度英雄鏡頭。兩個角色映襯著真實的日落天空。風勢增強。屋頂的警示燈柔和閃爍。

鏡頭 6 (7.5-9s):
兩個快速特寫：先是角色 A 的眼睛，再來是角色 B 的眼睛。不要分割畫面，不要出現多餘的臉。

鏡頭 7 (9-10.5s):
動態環繞式雙人鏡頭。鏡頭繞著兩人移動，他們緩慢地背對背擦身而過，但不接觸。不要出現無人機。

鏡頭 8 (10.5-12s):
從高處俯拍直升機停機坪的圓形區域。兩個角色分開站立，面朝不同方向。潮濕的混凝土只反射抽象的天空和城市燈光。

鏡頭 9 (12-13.5s):
屋頂邊緣附近的情緒化雙人中景。角色 A 看著城市。角色 B 稍微站在她身後，接著轉身面向天際線。

鏡頭 10

This video is a 15-second emotional music-video style sequence in a fully live-action photorealistic rooftop world.

Use {{Image 1}}  only as Character A identity reference.
Use {{Image 2}}  only as Character B identity reference.
Use the images only for face, hairstyle, outfit, accessories, colors, and silhouette.

Only Character A and Character B are anime-style.
The rooftop, sky, city, lights, props, reflections, camera optics, and atmosphere must look like real live-action footage.
Do not copy the anime style from 「アット」Image1 or 「アット」Image2 into the background.

Character rules:
Exactly two characters only: Character A and Character B.
No third person, no clones, no duplicate characters, no background people, no reflection-person, no shadow-person.
Do not turn the characters into realistic humans.
Keep both characters on-model, with stable faces, outfits, hair, and accessories.

Environment:
A realistic urban rooftop at sunset shifting into blue hour.
Wet concrete, helipad markings, metal railings, rooftop door, pipes, air conditioners, utility boxes, red warning lights, puddles, distant city lights, and a real skyline.
Strong wind moves hair, ribbons, sleeves, skirts, and loose fabric.

Strict avoid:
No anime background.
No illustrated rooftop.
No painted sky.
No cel-shaded buildings.
No visible drone.
No text overlays.
No logos.
No extra limbs.
No costume changes.
No face distortion.

Shot 1 (0-1.5s):
Wide rooftop opening. Character A stands near the railing left, Character B stands farther back right. Low camera glides forward over wet concrete and helipad markings.

Shot 2 (1.5-3s):
Close-medium on Character A. She turns her head toward camera on the beat. Hair and ribbon move in the wind. Background stays photorealistic and blurred.

Shot 3 (3-4.5s):
Close-medium on Character B. She looks down, then raises her eyes toward Character A. Sunset rim light outlines her hair.

Shot 4 (4.5-6s):
Medium-wide two-shot. Character A left, Character B right, both already visible. Character B takes one small step toward Character A. Camera tracks sideways. Exactly two characters.

Shot 5 (6-7.5s):
Low-angle hero shot. Both characters against the real sunset sky. Wind intensifies. Rooftop warning lights blink softly.

Shot 6 (7.5-9s):
Two quick close-ups: Character A’s eyes, then Character B’s eyes. No split screen, no extra faces.

Shot 7 (9-10.5s):
Dynamic orbit-like two-shot. The camera circles them as they slowly pass back-to-back without touching. No drone visible.

Shot 8 (10.5-12s):
High-angle shot over the helipad circle. Both characters stand apart, facing different directions. Wet concrete reflects only abstract sky and city lights.

Shot 9 (12-13.5s):
Emotional medium two-shot near the rooftop edge. Character A looks at the city. Character B stands slightly behind, then turns toward the skyline.

Shot 10

• 提示詞為日文：直接貼上即可，模型能理解。