FEE BABY 時尚女裝 FEE BABY 首頁
BL12

時間暫停

AI Reels 教學 · Taipei Main Station
影片中拍攝者在台北車站月台捏主角臉頰的畫面

按下碼錶,台北車站月台上所有人瞬間定格,只有拿鏡頭的人還能動:繞著她走一圈,最後偷捏一下她的臉,再按一次碼錶,時間恢復。

這支 25 秒的直式影片是用 AI 生成的。下面是我們實際做出成片的完整流程和提示詞,換上你自己的主角和穿搭就能做出同款。

看完整影片(Instagram)

成品分鏡

STEP 1準備照片

STEP 2生成設定

工具
Higgsfield 的 Seedance 2.5
模式
全能參考(Omni reference),上傳 STEP 1 的照片當參考圖
長度
25 秒
比例
9:16 直式
音訊
開啟(月台環境音和碼錶聲由 AI 生成,配樂最後再加)

STEP 3貼上提示詞

綠色底的地方換成你自己的內容,其他照抄就好。

中文完整版

寫實的手機第一人稱錄影,直式 9:16,一鏡到底、中間不剪接的手持拍攝,像 iPhone 錄的影片,現場是月台的日光燈,帶一點手持晃動。輕鬆、俏皮、可愛的時間暫停整人影片。 場景:真實的台北捷運台北車站板南線(藍線)地下月台,和開場圖一致:全高式玻璃月台門、門上方藍色色帶、灰色石材地板與排隊標線。整支影片唯一能看清楚的字,是和開場圖相同的站名標示(台北車站 Taipei Main Station),每次出現都要完全正確;其他招牌、螢幕和廣告一律模糊看不清。不要字幕、不要浮水印、不要 Logo。 主角:和參考照片是同一個人,臉、眼睛、眉毛、妝容、膚色、深棕色長髮與分線都一樣;二十多歲的成年女生,自然睜開眼睛。穿著和全身參考照完全相同:敞開的米白色皮革襯衫外套(白色釦子、胸前口袋)、白色細肩帶背心、米白色鬆緊腰短褲、白色短襪、奶茶色麂皮堆堆短靴,右手提白色小包。不論從哪個角度看,她的臉、頭髮和衣服全程都不能改變。 拍攝者:畫面中只看得到他的右手(男生的手、深色袖子)。 0–4 秒:列車已經停妥、門打開。她從車廂走上月台,帶著淺淺的微笑自然地朝鏡頭走來,旁邊有通勤的乘客在走動。 4–6 秒:拍攝者把銀色機械碼錶舉到畫面前景(像碼錶參考圖那樣),用拇指按下上方的按鈕。所有東西瞬間定格:她停在走路的半步、一隻腳在前,幾縷髮絲停在半空中,每個乘客都像雕像一樣停在動作中途。手把碼錶放下、移出畫面。 6–8 秒:只有鏡頭還能動,穿過定格的乘客走向她。其他人完全不動,也不眨眼。 8–17 秒:鏡頭在距離她約 1.5 公尺的地方慢慢繞她完整一圈,始終讓她在畫面中央、全身入鏡:正面,約 10 秒到她的左側,約 13 秒到她背後(看得到長髮和外套背面),約 15 秒到右側,約 17 秒回到正面。她全程維持同一個走路姿勢,完全不動。 17–22 秒:鏡頭推近到她胸口以上的臉部近景(像捏臉參考圖那樣)。他的右手伸進畫面,用拇指和食指輕輕捏住她的右臉頰:臉頰皮膚被擠壓、柔軟地鼓起,有真實的彈性和皮膚紋理,放開後自然彈回。她的表情仍然定格。 22–25 秒:他再按一次碼錶,時間恢復:乘客繼續走動,她眨眨眼、一臉疑惑,用指尖摸摸右臉頰,然後看向鏡頭,害羞又覺得好笑地笑出來。 整體清新單純,鏡頭不要停留在她的身體上。 聲音:真實的月台環境音,有人聲和車門提示音;5 秒按碼錶時清脆地喀一聲,接著環境音幾乎消失,只剩輕輕的時鐘滴答聲和拍攝者的腳步聲;22 秒第二次按碼錶後,車站環境音回來。不要配樂。不要對白、旁白或歌聲;她嘴巴閉著,全程不說話。

英文原版(我們實際生成成片用的版本)
English

Photorealistic smartphone POV footage, vertical 9:16, ONE continuous handheld take with no cuts, natural iPhone video look, real fluorescent platform lighting, subtle handheld sway. Lighthearted, playful, cute time-stop prank video. Location: the real Taipei Metro Taipei Main Station, Bannan line (blue line) underground platform, exactly as in the start image: full-height glass platform screen doors, blue line-colour band above the doors, grey stone floor with queue markings. The only readable text in the whole video is the station name sign exactly as in the start image (台北車站 / Taipei Main Station); keep it identical and correctly spelled whenever visible. All other signs, screens and ads stay blurred and unreadable. No subtitles, no captions, no watermark, no logos. The woman is exactly the person in the reference photos: same face, eyes, eyebrows, makeup, skin tone, long dark-brown hair and parting. She is an adult in her mid-twenties with her eyes naturally open. Same outfit as the full-body reference photo: open cream faux-leather oversized shirt jacket with white buttons and a chest pocket, white camisole, cream elastic-waist shorts, white crew socks, beige suede slouch ankle boots, small white handbag in her right hand. Her face, hair and outfit must stay identical from every angle for the whole video. The camera operator is never seen except his right hand (a man's hand, dark sleeve). 0-4s: The train has stopped and the doors are open. She steps off the train onto the platform and walks naturally toward the camera with a soft smile, commuters walking around her. 4-6s: The camera operator raises a silver mechanical stopwatch into the foreground, exactly like the stopwatch reference, and presses its top button with his thumb. Instantly everything freezes: she stops mid-stride with one foot forward, strands of her hair suspended in mid-air, every commuter frozen mid-motion like a statue. The hand lowers the stopwatch out of frame. 6-8s: Only the camera keeps moving. It walks forward between the frozen commuters toward her. Nobody else moves or blinks. 8-17s: The camera slowly orbits a full 360 degrees around her at about 1.5 metres, always keeping her centred, full-body framing: front, then her left side profile at about 10s, then her back at about 13s showing her long hair and the back of the jacket, then her right side at about 15s, then back to the front at about 17s. She stays perfectly frozen in the same mid-stride pose the whole time. 17-22s: The camera pushes in to a chest-up close-up of her face, like the cheek-pinch reference. His right hand reaches in and gently pinches her right cheek between thumb and index finger: the cheek skin compresses and bulges softly with realistic elasticity and skin texture, then springs back naturally when he lets go. Her expression stays frozen. 22-25s: He clicks the stopwatch again and time resumes: the commuters continue walking, she blinks, looks puzzled, touches her right cheek with her fingertips, then looks into the camera and breaks into a shy, amused smile. Tasteful and innocent, no lingering on her body. Audio: realistic platform ambience with crowd murmur and the train door chime; a crisp stopwatch click at 5s, then all ambient sound cuts to near silence with only a soft clock ticking and the operator's quiet footsteps; a second click at 22s and the station ambience returns. No music. No dialogue, no voiceover, no singing; her lips stay closed and she does not speak.

STEP 4開場畫面(建議)

丟給任何能附參考圖的 AI 生圖工具,附上臉部特寫和全身照,生成後放在影片的第一格:

開場畫面用

寫實照片,像 iPhone 錄影暫停下來的一格,直式 9:16。地點是台北捷運台北車站的板南線地下月台:全高式玻璃月台門、上方藍色板南線色帶、淺灰色石材地板。列車剛停妥、門打開,參考照中的女生(臉、妝容、髮型與參考照完全相同,自然睜眼)剛下車,朝鏡頭走來、淡淡微笑,全身入鏡在畫面中央。服裝與全身參考照完全相同。她身後上方的站名標示清楚寫著台北車站與 Taipei Main Station,字要完全正確;其他文字一律模糊看不清。現場日光燈、手機錄影質感、細微雜訊,不要浮水印或 Logo。

STEP 5先出草稿,再定稿

先用 480p 草稿一次出 2 版,挑表演最好的那版,再轉成 1080p 定稿,比直接出 1080p 省點數。我們 2026 年 10 月實際的花費:25 秒草稿每版 75 點、1080p 定稿 300 點,定稿大約等了 40 分鐘。

草稿要逐格看這幾件事:

STEP 6配樂與發文

避坑技巧

肖像權提醒:只用你自己的照片,或已經取得本人同意的照片。不要拿名人、網紅或陌生人的照片來做。

影片裡的穿搭