想把自己變成童年電影裡的主角?這篇整理 Dream 夢幻復古短片的製作流程:先建立固定角色,再製作對嘴/動作片段與四個回憶場景,最後剪輯成完整影片。下方保留各場景的英文提示詞,可以依照自己的角色與故事修改。
工具依原始分享文件:角色圖使用 Nano Banana Pro 或 Seedream 5;影片使用 Seedance 2.5。不同平台的入口與可用功能可能不同,請先確認所用平台支援需要的參考素材。本文未取得原作者的生成秒數、比例與解析度設定;以下流程與檢查建議供操作參考,不代表已實測重現相同結果。
開始前:準備素材與製作順序
準備 2–3 張自己的清楚照片,用來建立童年版角色圖;若要做對嘴/動作模仿片段,另備一段作為表情、口型與動作依據的來源影片。各場景提示詞提到的構圖或色調參考圖,也要一併準備,並確認自己有權使用。
製作順序:角色圖 → 對嘴/動作片段(依需要製作)→ S1 草地與羊群 → S2 躺草地看雲 → S3 騎車與白貓 → S4 放紙船 → 剪輯與輸出。四個場景請分開生成與儲存。
步驟 1|建立並下載童年版角色圖
1. 開啟 Nano Banana Pro 或 Seedream 5 的圖片生成/編輯介面,上傳 2–3 張自己的照片作為人物參考。
2. 明確告訴模型:「以我上傳的人物照片為參考,製作同一人物的童年版本角色設定圖,呈現正面、側面與背面,並維持一致的臉部特徵與服裝。」再接上下方的服裝與髮型提示詞。下方英文主要描述造型,童年化與多視角需求要另外說明。
3. 生成後先檢查:不同角度是否是同一個女孩、衣服和髮型是否一致、臉與四肢有沒有明顯異常。挑選滿意的一張下載,後續場景都使用這張固定角色圖。

Charactersheet girl is wearing A charming, elegant black-and-cream vintage-inspired children’s outfit, with a soft minimalist Korean/Japanese fashion aesthetic.
CLOTHING: A loose, slightly cropped black collarless jacket/cardigan made from soft matte fabric, featuring a clean round neckline, subtle small black buttons down the center front, relaxed dropped shoulders, and wide long sleeves. The jacket has a gently structured, boxy silhouette with a slightly flared hem.
Underneath, a warm ivory/cream blouse with a soft flowing shape. The blouse extends below the black jacket and finishes with a delicate scalloped cream lace trim, creating a visible romantic lace layer around the waist. Keep the cream fabric slightly textured and lightweight rather than bright white.
Pair it with very loose black balloon/harem trousers, dramatically roomy through the hips and legs with a soft rounded silhouette, tapering gently toward the ankles. The fabric should look lightweight, matte, comfortable, and naturally gathered, creating subtle folds and volume.
Finish with simple black rounded-toe Mary Jane flats, minimal design, matte finish, with a small strap across the top of each foot. No socks, no jewelry, no bag, no additional accessories.
HAIRSTYLE: Dark brown to nearly black hair styled into a soft, slightly messy low bun at the back, with the hair naturally gathered rather than tightly slicked. Keep a soft center/near-center part, with long wispy curtain-like strands and fine flyaway hairs framing both sides of the face. Several loose strands fall naturally around the temples, cheeks, and ears. Add a tiny cream-colored flower hair clip positioned subtly near the side/back of the bun. The overall hairstyle should feel delicate, effortless, slightly tousled, natural, and youthful.
步驟 2|製作對嘴/動作模仿片段(選用)
這段提示詞除了背景,也要求模仿來源影片的口型、表情與動作,因此需要「角色圖+來源影片」。若只想製作下方四個回憶場景,可以先跳到步驟 3。
1. 在使用平台中選擇能接收角色圖片與來源影片的生成流程,確認模型與素材支援情況;原稿標示影片工具為 Seedance 2.5,但沒有交代平台內的具體模式名稱。
2. 上傳步驟 1 的角色圖與來源影片,分別指定為人物外觀與表演動作的參考,再貼上下方提示詞。
3. 檢查淡黃色房屋與花園是否符合需求;若要改背景,就替換提示詞中描述房屋與花園的句子。生成後檢查口型、表情與動作的時間是否對得上,並下載可用片段。只有背景圖片、沒有來源影片時,不適合直接照貼這段動作模仿指令。
補充|用 Genjutsu 做人物替換,會花多少 credits?
如果你要保留來源影片的動作、口型或場景,再換成自己的角色,可以在 Higgsfield 使用 Genjutsu。Motion Transfer 著重轉移表演動作,Object Swap 著重替換指定人物或物件。這一節只說明 Genjutsu 的費用,不代表下方 S1–S4 的 Seedance 場景生成也使用相同費率。
Genjutsu 是依來源影片長度計價。先把要測試的來源影片剪短,再上傳詢價;不要以為最後只剪出 4 秒,就只會付 4 秒的費用。每次重新生成都要重新計算預算,最終以你當下選定模型、來源素材與畫質後,Generate 按鈕顯示的 credits 為準。
單次生成:今天查到的報價範例
以下依 2026/10/20 已有的來源影片與設定,透過 Higgsfield 成本預估工具重新詢價;詢價未啟動生成。480p Motion Transfer 範例使用關閉生成音訊、高位元率;Object Swap 使用關閉生成音訊。這些是指定設定的報價範例,不能直接套用到其他畫質或設定。
| Genjutsu 模式 | 來源片長 | 畫質 | 單次預估 |
|---|---|---|---|
| Motion Transfer | 4 秒 | 480p | 12 credits |
| Motion Transfer | 8 秒 | 480p | 24 credits |
| Motion Transfer | 20 秒 | 480p | 60 credits |
| Motion Transfer | 29 秒 | 480p | 87 credits |
| Object Swap | 5 秒 | 480p | 15 credits |
| Object Swap | 6 秒 | 480p | 18 credits |
| Object Swap | 6 秒 | 1080p | 120 credits |
| Object Swap | 8 秒 | 1080p | 160 credits |
以同為 6 秒的 Object Swap 為例,這次查到的報價為 480p 每次 18 credits、1080p 每次 120 credits。若各生成 3 次,就是 54 與 360 credits。成片雖然只有幾秒,多次試臉、服裝與表情,點數仍會一路累加。
一天測試下來,點數怎麼累積?
以 2026/10/20 台灣時間截至本次查核的帳戶生成紀錄為例,共有 17 支已完成的 Genjutsu 影片:9 支 Motion Transfer(480p),依目前報價回估合計 330 credits;4 支 Object Swap(480p),合計 66 credits;4 支 Object Swap(1080p),合計 520 credits。這 17 支影片合計回估約 916 credits,其中僅 4 支 1080p Object Swap 就占約 57%。
注意:916 credits 是依當下報價回估的測試規模,不是經逐筆扣點帳單核實的實際支出,也不是完成這篇 Dream 教學的固定成本。這份統計包含當天不同 Genjutsu 測試,未把它們全部視為本片製作;角色圖及其他圖片生成另計。另有 1 筆未完成的 Genjutsu 記錄未納入,退款、免費次數或方案優惠也未做帳務核銷。
怎麼測,才比較不容易燒光點數?
1. 先裁來源影片:挑 4–6 秒、臉部清楚而且包含關鍵動作的一小段測試。29 秒的來源即使只測一次,這次 480p Motion Transfer 報價也要 87 credits。
2. 先用 480p 確認角色、服裝與動作方向,再考慮用相同素材及提示詞重新生成 1080p。高畫質重生會再次花點數,結果也可能不同;這不是免費把原成品升級。
3. 一次只調整一個重點,例如先固定臉,再調服裝或背景,避免每次同時改太多而無法判斷問題。生成前先檢查來源影片與角色參考圖是否選對。
4. 預算用「單次報價 × 預計生成次數」計算。例如 6 秒 480p Object Swap 先預留 3 次,約 54 credits;若接著重生 1 次 1080p,再加約 120,合計約 174 credits。這只是預算例子,不保證三次內得到滿意結果。
5. 遇到失敗或退款提示,請查看實際退點狀態;看到一次 Generate 報價,不能直接視為一次成功扣點。開始下一輪前,先核對目前剩餘 credits。
步驟 3|逐一生成四個回憶場景
每次先建立一個獨立生成任務,上傳固定角色圖,再加入該場景需要的構圖/色調參考圖,最後貼上對應的英文提示詞。提示詞中的「same girl」「first reference image」不代表模型會自動記得上一個任務;每次都要重新提供所需素材,並確認平台如何指定參考用途。
先決定成片要直式還是橫式,並讓所有片段保持一致。建議先用平台可用的較短片長試做、檢查人物與動作後再調整;實際秒數、畫質與素材數量以平台提供的選項為準。下方多鏡頭提示詞若未能完整執行,可以拆成獨立鏡頭生成,再到剪輯軟體串接。
S1|女孩與羊群在草地奔跑
素材:固定角色圖+草地奔跑的構圖/色調參考圖。先確認哪張圖用來固定人物、哪張用來參考暖色復古氛圍,再貼上下方 S1 提示詞生成。
檢查重點:女孩是否仍是同一角色、羊群有沒有重疊或肢體變形、鏡頭是否從身後跟拍,以及草地是否維持柔和的橄欖綠。滿意後下載並命名為 S1。


Create a nostalgic, playful outdoor scene of the character-sheet girl running freely through a huge open grassy field while playing with a group of small, extremely fluffy sheep.
COLOR GRADE & FILM LOOK — STRICTLY LOCKED TO REFERENCE
Match the exact visual mood and color treatment of the provided reference image. The footage should feel like a warm, slightly faded piece of vintage film rather than a modern digital video.
Use warm late-afternoon sunlight with a soft golden/amber cast. Greens should be muted, earthy olive-green, never bright or neon. Skin tones remain naturally warm and slightly creamy. Highlights have a subtle yellow-golden warmth, while shadows are soft, slightly brown/green, and never crushed to pure black.
Keep moderately low saturation, restrained contrast, gently lifted blacks, slightly faded colors, soft highlight roll-off, and subtle atmospheric warmth. Add fine organic film grain, very subtle analog noise, slight softness, mild natural halation around strongly illuminated edges, and a tiny amount of vintage lens haze. The image should have the imperfect texture of an older 16mm/35mm family-film photograph mixed with an early-2000s cinematic home-video feeling.
Do not use cool cyan grading, strong blues, teal-and-orange grading, HDR, deep crushed blacks, extremely sharp digital details, oversaturated grass, glossy commercial lighting, or clean modern smartphone-video processing.
ENVIRONMENT
The girl is running through an enormous, peaceful field of soft rolling green grass, extending toward a simple distant horizon. Keep the landscape minimal and natural. The warm sunlight falls sideways across the grass, producing long, very soft shadows and beautiful patches of golden illumination.
No houses, roads, fences, mountains, vehicles, crowds, poles, or modern objects. Only the girl, sheep, grass, distant rolling terrain, and natural sky.
SHEEP — VERY FLUFFY
Around her is a small flock of adorable, relatively small sheep with an unusually huge amount of wool. Their bodies should appear round and extremely fluffy because of their dense, long, soft, slightly messy wool. Think tiny sheep surrounded by oversized cloud-like coats of natural wool.
They are not ordinary thin-wool sheep. Make them noticeably extra fluffy, puffy, fuzzy and soft, while maintaining completely realistic sheep anatomy.
Several sheep run directly beside her, some slightly ahead, and several behind her. Their fluffy wool naturally bounces and jiggles as they run, while their little legs remain visible underneath.
ACTION & PERFORMANCE
The girl runs freely through the grass laughing and genuinely playing with the sheep. She isn't simply jogging while the animals follow her.
She playfully runs between them, changes direction unexpectedly, reaches toward one fluffy sheep running beside her, pulls her hand back while laughing, then continues running. One or two sheep playfully race ahead of her while the others follow around her.
Her performance is extremely natural, spontaneous and human. Her smile develops naturally rather than remaining frozen. She laughs, catches her breath, looks toward the sheep, briefly looks back toward the camera, and then turns forward again.
Her hair moves naturally from her running and the gentle breeze. Her clothes react realistically to every step.
CAMERA
Start from behind the character-sheet girl, approximately at waist/chest height, similar to the original running-field reference.
The camera operator is physically running behind her, creating subtle organic handheld movement. Keep her generally near the center of frame but don't artificially lock her there. Allow tiny framing mistakes and natural vertical movement from footsteps.
As she runs, some fluffy sheep occasionally enter closer to the camera and cross through the foreground while she continues farther ahead.
Keep it intimate and observational—as though a friend happened to capture this beautiful moment on an old camera.
No drone shot. No orbit. No gimbal-perfect stabilization. No dramatic zoom. No artificial camera movement. No slow motion.
LIGHTING
Warm natural late-afternoon sunlight, matching the reference image. Soft golden sunlight should catch the edges of her hair, shoulders, grass, and the sheep's thick wool. The sheep's wool should pick up beautiful creamy golden highlights while retaining slightly darker warm shadows between individual wool strands.
The lighting should feel completely natural and available, not artificially staged.
SOUND
NO MUSIC.
Only natural location audio: her genuine laughter, slightly breathless breathing while running, feet moving through grass, sheep's little footsteps, occasional soft sheep sounds, thick wool brushing against grass, grass rustling, and a gentle natural breeze.
FINAL FEELING
The entire scene should feel like a forgotten childhood memory captured on warm vintage film—quietly cinematic, imperfect, innocent, playful, nostalgic, dreamy and emotionally warm.
STRICT NEGATIVE: no modern glossy grading, no cold/cyan cast, no vibrant emerald-green grass, no oversaturation, no HDR, no excessive contrast, no crushed blacks, no perfect digital sharpness, no artificial studio lighting, no slow motion, no duplicated sheep, no fused sheep, no malformed legs, no floating animals, no unnatural running animation, no robotic girl movement, no exaggerated acting, no text, and no music.
S2|躺在草地看雲,牛形雲朵慢慢散開
素材:固定角色圖+與 S1 一致的色調參考圖。貼上下方 S2 提示詞,這段包含「女孩躺在草地仰望」與「天空主觀視角」兩個鏡頭。
檢查重點:先看到女孩看向天空,再切到雲朵;牛的輪廓應由雲形成,不能變成天空中真的牛。如果兩個鏡頭無法在一次生成中完成,建議將人物鏡頭與雲朵鏡頭拆開製作,再剪接;命名為 S2-A、S2-B。
SCENE 1 — LYING IN THE FIELD, LOOKING AT THE SKY
The girl is now lying completely relaxed on her back in the same huge open grassy field from the earlier running scene. The field consists of soft, slightly uneven rolling green grass with a simple distant horizon. No buildings, roads, fences, crowds, mountains, or modern objects.
She lies directly in the grass, comfortably settled into it. Her body is relaxed and still. Both hands rest naturally underneath the back of her head, with her elbows opened loosely outward, like someone resting in a field and watching clouds pass overhead.
Her legs are relaxed rather than perfectly straight—one knee can be slightly bent while the other leg rests naturally against the grass.
She is looking straight upward toward the sky, completely absorbed in watching the clouds.
Her expression is peaceful and innocent, with a very subtle sense of curiosity. She isn't posing or looking toward the camera. Her eyes slowly follow something moving above her. After a few seconds, the smallest soft smile appears on her face, as though she has recognized a funny shape in the clouds.
A gentle breeze softly moves her hair, loose strands around her face, her clothing, and the tips of the surrounding grass.
CAMERA: begin with an intimate side/three-quarter medium shot from very low in the grass, approximately at the level of her body. Some blades of grass naturally fall softly out of focus in the foreground. Her face is visible in profile/three-quarter profile as she stares upward.
Then hold on her eyes looking toward the sky for a moment.
Do not show what she sees yet.
Natural handheld imperfections only. No orbit, no dramatic push-in, no zoom, no slow motion.
CUT TO — HER POV OF THE SKY
Make a clean natural cut from her looking upward to a full-frame view of the sky, as though we are now seeing exactly what she sees.
The entire frame becomes a huge dreamy sky filled with many soft, layered clouds.
The sky and clouds have a subtle dusty pink / faded peach / pale rose tint, but keep it restrained and natural within the established vintage palette. It should not become bright fantasy pink or purple.
There are many different layers of clouds—some thin and wispy, some larger and softly illuminated, with gentle variations of faded pink, warm ivory, dusty peach, muted gray and pale cream.
Among the clouds is one unusual formation that naturally resembles the silhouette of a cow.
The cow shape must be made entirely from a real cloud. It should not look like a cartoon cow, illustration, CGI animal, or perfectly sculpted cow. It is simply an accidental cloud formation where the viewer can recognize a cow-like body, small head, four vague legs, and a soft suggestion of a tail.
Keep the shape slightly ambiguous, like real cloud watching—you look at it and think, that looks like a cow.
CLOUD TRANSFORMATION
Hold on the cow-shaped cloud for a brief moment.
The wind then slowly begins changing its shape naturally.
First, the edges around the legs become softer and start drifting apart. Then the head gradually stretches into the surrounding cloud. The body loses its recognizable outline as wispy pieces gently separate.
Within a few moments, the cow shape completely dissolves and spreads apart into ordinary clouds.
It should feel like genuine atmospheric cloud movement—not a magical transformation or morphing VFX.
The separated wisps continue drifting peacefully across the frame.
MASTER 90s COLOR GRADE — STRICTLY LOCKED
Match the first reference image's color character across the girl, grass, and sky.
Authentic 90s film/home-video memory aesthetic: slightly warm and faded, restrained saturation, creamy skin tones, earthy muted olive-green grass, soft yellow/amber warmth in highlights, slightly green-brown shadows, gently lifted blacks, low-to-moderate contrast, and softly rolled-off highlights.
For the sky shot, introduce the dusty pink/peach cloud tones while keeping the same faded 90s base. The pink should feel aged and sun-faded rather than digitally saturated.
Add fine organic grain, subtle analog noise, slight softness, mild color bleed, faint halation, gentle haze, imperfect exposure, and a tiny amount of natural flicker.
The footage should look like an actual memory captured in the 1990s, not modern 4K footage with a retro filter.
AUDIO — NO MUSIC: gentle wind moving through the grass in the first shot, distant birds, subtle fabric and grass rustling. When cutting to the sky, keep only the soft wind and distant natural ambience.
STRICT NEGATIVE: no cartoon cow, no actual cow floating in the sky, no perfectly detailed cow-shaped cloud, no magical glowing cloud, no fantasy VFX, no rapid morphing, no dramatic sunset, no neon pink or purple sky, no HDR, no cyan/teal grade, no oversaturated grass, no modern digital sharpness, no dramatic camera movement, no slow motion, no dialogue, no text, and no music.
S3|女孩騎腳踏車,後座載著白貓
素材:固定角色圖+高角度騎車構圖參考;若要固定白貓外觀,也需提供白貓參考圖。提示詞提到「同一隻白貓」,但前面尚未建立白貓角色,請在這裡先選定外觀。
貼上下方 S3 提示詞生成。檢查攝影機是否維持在女孩上方、白貓是否待在後座籃內,以及踩踏、車輪與手握車把的動作是否自然。滿意後下載並命名為 S3。
S4|女孩放下紙船,切到水面特寫
素材:固定角色圖+池塘邊蹲下放船的構圖參考圖,並延續前面相同的暖色調。貼上下方 S4 提示詞,依序呈現女孩放船與水面紙船特寫。
檢查重點:紙船是否從手中自然接觸水面、鬆手後才開始漂浮、切到特寫後是否仍是同一艘紙船。如果模型沒有完成切鏡,建議分成 S4-A 放船、S4-B 紙船特寫兩段生成,再在剪輯時接起來。
SCENE 1 — GIRL PLACING THE PAPER BOAT
Create a peaceful outdoor scene inspired by the supplied reference composition. The girl is crouching/squatting naturally at the edge of a very small, shallow pond surrounded entirely by uneven green grass.
Keep the camera at approximately the same elevated wide three-quarter angle as the reference, looking slightly downward toward both the girl and the pond. The pond occupies a large portion of the frame, allowing her reflection and the surrounding landscape to appear naturally on its surface.
She is alone.
The girl is holding a small handmade white paper boat carefully between her fingers.
She leans forward over the edge of the pond, extending one hand slowly toward the water while her other hand naturally helps her balance against her knee.
Her expression is quiet, gentle and slightly thoughtful. She watches the little paper boat carefully.
She slowly lowers the white paper boat until its bottom gently touches the surface of the water.
For a moment she still holds it with her fingertips.
Then she carefully releases it.
The boat floats.
Her fingers remain close to the water for a second as she watches it begin to move away. Tiny circular ripples spread outward from where the boat touched the surface.
Keep her movements small, delicate and completely natural—no dramatic gestures.
The still pond naturally reflects portions of her silhouette, the grass, trees and warm sky.
CUT TO SCENE 2 — CLOSE-UP OF THE PAPER BOAT
Cut immediately as she releases the boat.
Change to an intimate low close-up at almost water level, focused entirely on the same small white paper boat floating on the pond.
At the beginning of this shot, her fingertips are still visible gently releasing the top/back edge of the boat.
Her fingers slowly pull away and leave the frame.
The camera remains focused on the paper boat.
The boat makes a tiny natural rocking movement from being placed into the water. Small ripples travel outward around it.
Then a very gentle current/breeze begins carrying the boat slowly forward across the pond.
Do not make the boat travel unnaturally fast. It should quietly drift.
Show subtle reflections moving underneath it—the faded sky, dark shapes of distant trees and muted green grass along the shoreline.
Keep the camera very close to the surface, making the tiny paper boat feel emotionally important.
No hands remain in frame after she releases it.
CAMERA CONTINUITY
Shot 1: elevated wide three-quarter composition matching the supplied reference—girl crouched beside the pond, surrounding grass visible, reflection visible.
Shot 2: hard natural cut to a tight water-level close-up precisely when she places/releases the boat.
Do not slowly move the first camera down toward the boat. It should feel like two separate shots from an old movie.
MASTER COLOR GRADE — LOCKED
Use the exact same warm nostalgic grade established by the sheep-field reference.
Warm late-afternoon 90s-film character, creamy natural skin, muted olive and moss greens, soft golden/amber sunlight, warm neutral shadows, slightly faded saturation, gently lifted blacks, restrained contrast, smooth highlight roll-off, fine organic grain, subtle analog noise, mild vintage lens softness, delicate halation and naturally imperfect exposure.
The pond should be a dark muted gray-green, reflecting the environment naturally—not bright blue.
The white paper boat should appear slightly warm ivory/creamy white under the sunlight, rather than sterile digital white.
Keep the whole image soft, tactile and slightly imperfect. No modern HDR, no teal/cyan grading, no neon greens, no ultra-sharp digital appearance.
SOUND — ABSOLUTELY NO MUSIC
NO MUSIC. NO SCORE. NO INSTRUMENTAL SOUND.
Only natural ambience: gentle wind through the grass, distant birds, subtle insects, tiny movements of water against the bank, her clothes softly rustling as she crouches, the faint touch of paper against water, and delicate water sounds as the boat begins floating.
No dialogue.
The feeling should be quiet, innocent, nostalgic and slightly bittersweet—like a tiny forgotten moment captured on film in the 90s.
步驟 4|剪輯、檢查與輸出
1. 將下載的片段匯入慣用的剪輯軟體,依 S1 → S2 → S3 → S4 排列;對嘴/動作片段可依你的故事安排在開頭或其他位置。
2. 剪掉人物變形、動作失誤與多餘等待的部分。S2 在女孩仰望後切到天空;S4 在放開紙船的動作附近切到水面特寫。
3. 對照各段的人物外觀、畫面比例與色調,必要時在剪輯軟體微調,避免相鄰鏡頭顏色突然改變。原提示詞要求不生成音樂,只保留環境音;請逐段聽過,處理突兀的音量與接點。
4. 建議依發布平台需求輸出 MP4,輸出後完整播放一次,確認畫面沒有黑格、錯序或變形片段,音訊也銜接自然。
常見問題:人物換臉時,先確認每段都附上同一張角色圖;構圖或色調對不上時,補上提示詞所指的參考圖;動作太多、切鏡失敗時,減少單次任務的內容或拆成多段。提示詞中的「STRICT」「exactly」是要求,仍需人工檢查生成結果。