照片變環繞運鏡效果｜DAY 8 完整提示詞與製作教學
準備人物照片、生成多角度場景，再用 AI 做出環繞運鏡。附可直接複製的場景與影片提示詞，以及避免畫面翻轉的調整方法。
Brayden 布萊登
10/1/20262 分钟阅读
《從零開始學短影片行銷》DAY 8
只有一張人物照片，也想做出鏡頭繞著你移動的效果？這次用戶外 Vintage Market 的照片示範：先補出不同角度的場景，再讓 AI 生成帶有前後推拉、高低變化的環繞運鏡。
關鍵是把人物、場景、鏡頭路線分開交代。下面整理成三步，附上可以直接複製的提示詞。
① 準備一張清楚的人物照片
選擇光線充足、臉部清楚、服裝完整的照片。想展示全身穿搭，就讓鞋子也出現在畫面中，四周留一點空間。盡量避免手腳被遮住，或人物太小。
這張照片負責告訴 AI「主角長什麼樣子」。範例是戴墨鏡、穿彩色賽車外套的人物；換成你的照片時，也要同步修改後面提示詞中的外觀與場景描述。
② 用 ChatGPT 補出同一個場景的不同角度
把人物照片上傳到 ChatGPT，使用圖片生成功能，貼上以下提示詞。目標是一張五格的場景參考圖：上方一張廣角，下方四張不同視角。
這一步是讓 AI 推想照片沒有拍到的區域，並非把現場精確掃描成 3D。生成後先檢查帳篷、衣架、走道與光線是否看起來屬於同一個地方。
多角度場景提示詞｜複製以下全文
Use the uploaded portrait as the visual reference for the outdoor vintage market. Create a photorealistic five-panel environment reference sheet: one wide establishing view across the top, then four views in a 2-by-2 grid below.
Show the same location from the front, front-left, front-right, a slightly elevated view, and a low street-level view. Keep the blue and white tents, clothing racks, asphalt walkway, trees, sunny daylight and overall color palette consistent. Maintain a plausible, coherent layout between views.
Remove the main person from the source photo. Focus on the environment. Reconstruct unseen areas plausibly without introducing a different location. No captions, labels, logos added by the generator, or decorative frames. This is an environment reference sheet, not a sequence of different markets.
換場景的方法：把「outdoor vintage market」和帳篷、衣架等描述，改成照片中真正存在的環境。保留「同一地點、不同角度、光線一致」的要求。
③ 上傳人物＋場景，生成環繞運鏡
這次示範使用 DomoAI 的 Omni Reference 介面，選用 Seedance 2.5。將人物照片設為 Image1，多角度場景圖設為 Image2，選擇 9:16 直式、8 秒，再貼上影片提示詞。工具選項若有更新，以你實際看到的介面為準。
兩張圖片的任務不同：Image1 固定人物外觀，Image2 提供場景參考。提示詞也明確要求輸出完整畫面，避免 AI 把五格參考圖直接做成分割畫面。
環繞運鏡提示詞｜本次實際使用原文
Create an 8-second photorealistic vertical video, 9:16.
REFERENCE ROLES
Image1 defines the person's identity, hairstyle, sunglasses, outfit and body proportions. It is an appearance reference, not a required opening composition.
Image2 defines the vintage market environment only. Its panels show different views of one location. Render one full-screen scene, never a collage or split screen.
SUBJECT
The person stands in the same spot with feet planted, hands relaxed beside their thighs and a natural smile. Keep their body facing the original direction. Allow only subtle natural head and eye movement.
CAMERA SEQUENCE
0.0–1.0 seconds: Start close to the person's face, showing the head and shoulders. Quickly move backward to reveal their full outfit.
1.0–2.8 seconds: Move the camera sideways along a short curved path to a front-left three-quarter view. Rise slightly and look down toward the person. Keep the person prominent in frame.
2.8–4.6 seconds: Sweep back across the front toward a front-right three-quarter view. Lower the camera to waist height and aim slightly upward. The clothing racks and tents shift naturally behind the person as the camera changes position.
4.6–6.3 seconds: Make another brisk lateral arc back toward the front-left view at chest height. Use a quick acceleration followed by a brief readable pause.
6.3–8.0 seconds: Return to the front and move closer, ending with a head-and-shoulders portrait.
CAMERA ORIENTATION AND FRAMING
The camera physically travels around the person on short horizontal arcs. Keep the horizon level and the image upright throughout. Zero camera roll. The person's head always stays above their feet in the image.
In wide views, the person occupies approximately 60–80% of the frame height. Maintain enough distance to avoid cutting off the shoes.
Use a moderately wide lens with natural perspective, clear background parallax and brief motion blur during the fast movements.
CONTINUITY
Preserve the same person, clothing, market layout and sunny daylight throughout. Connect the camera movements into one continuous shot.
AVOID
No barrel rolls, upside-down frames, spinning image, full 360-degree rotation, extreme fisheye distortion, distant aerial views, tiny subject, rotating person, sliding feet, duplicate person, outfit changes or split screen.
我使用的工具是 DOMO AI。使用折扣碼 BRAYDEN 可以打九折喔～ 🎁
為什麼不只寫「360 度環繞」？
我們要的是攝影機移動、背景產生視差，但「旋轉」也可能被模型理解成整張畫面翻轉，或人物自己原地轉圈。因此這版先用短弧線左右移動＋高低變化＋推拉建立環繞感，不要求完整繞行一圈。
分秒描述是希望模型遵循的節奏，不代表每次都能精準執行。先把方向做對，再逐次增加移動幅度，每次只調整一個條件，比同時堆疊很多運鏡要求更容易判斷問題。
生成不像預期？先改這三個地方
畫面翻過來：保留「Zero camera roll」和「Keep the horizon level」，刪除完整 360 度旋轉的要求。
人物跟著轉圈：強調雙腳固定、身體保持原本朝向，移動的是攝影機。
人物變小或背景亂跳：縮短移動路線，保留人物佔畫面高度約 60–80% 的要求，重新檢查場景參考圖的一致性。
做好後，挑最順的一段放在短影片開場，接上原始照片與三步教學，觀眾就能同時看到成果和做法。
想收到更多 AI 實作教學？
訂閱我的電子報，繼續一起學 AI 工具、短影片行銷與內容創作。把看過的效果，變成自己也做得出來的作品。