생성형 영상 프롬프트 스킬

# 망고 에버랜드 릴스 — Flow 프롬프트

**선글라스 없는 버전입니다.** 망고 원래 얼굴로 갑니다.

## 0. 설정 요약

| 항목 | 값 |
| --- | --- |
| 모델 | Gemini Omni Flash 1.1 |
| 화면비 | 세로 9:16 |
| 샷 개수 | 5개 |
| 생성 길이 | 24초 |
| 최종 길이 | 12초 |

**왜 Omni Flash인가**

소재(Ingredients) 참조를 4초·6초 클립에 쓸 수 있는 모델이 이것뿐입니다.

Veo 3.1은 소재 참조가 8초에서만 됩니다.

음성 참조(`@Voice`)도 소재를 쓰는 생성에만 붙습니다.

---

## 1. 얼굴을 고정하는 방법

지난번에 얼굴이 매번 바뀐 이유는 셋이었습니다.

| 원인 | 이번 대책 |
| --- | --- |
| 선글라스 | 아예 뺌 |
| 컷마다 다른 설명 | 같은 문장 고정 |
| 유지 지시 없음 | keep 문장 추가 |

**선글라스를 뺐습니다**

AI가 매번 새로 그리던 물건이 사라집니다. 흔들림의 가장 큰 원인이었습니다.

대신 **분홍 하트 옷**이 망고의 표식입니다. 옷은 사진에 이미 있어서 새로 상상할 필요가 없습니다.

**망고 묘사 문장을 고정했습니다**

다섯 컷 모두 아래 문장을 글자 하나 안 바꾸고 씁니다.

```
@Mango, a small white Maltese dog with long straight white ear fur hanging past
its jaw, a short-clipped white body coat, large round dark brown eyes and a dark
brown nose, wearing a pink knit top with a white heart pattern and ruffled
shoulder trim
```

**유지 지시문을 넣었습니다**

각 컷 끝에 아래 문장이 들어갑니다.

```
Keep @Mango's face, ear fur and pink heart-patterned top exactly as they are in
the reference image and do not restyle, recolor or replace them.
```

**한 마리만 나오게**

각 컷 끝에 이 문장도 함께 들어갑니다.

```
Exactly one dog is in the scene and no second dog appears anywhere, not even in
the background.
```

---

## 2. 소재 이미지 만들기

### 어떤 사진을 쓰나

`20230225_131504.jpg`를 씁니다.

| 이유 | 설명 |
| --- | --- |
| 낮 빛 | 얼굴이 밝고 또렷함 |
| 전신 | 몸 비율을 읽을 수 있음 |
| 정면 얼굴 | 눈·코 위치가 보임 |
| 분홍 옷 | 캐릭터 표식이 됨 |

### 절차

1. Flow 프로젝트에 사진을 올립니다.
2. 이미지 생성(Nano Banana Pro)에 아래 프롬프트를 넣습니다.
3. 나온 이미지를 `Mango`로 저장합니다.
4. 모델명 → Video → Ingredients에 넣고 `@Mango`로 부릅니다.

### 소재 이미지 프롬프트

```
A small white Maltese dog identical to the one in the reference photo, with long
straight white ear fur hanging past its jaw, a short-clipped white body coat,
large round dark brown eyes and a dark brown nose, wearing the same pink knit top
with a white heart pattern and ruffled shoulder trim. Full body, standing, facing
the camera straight on with the head level. Plain light gray seamless background,
no clutter, no props. Soft even daylight, clean product-photo lighting, realistic
photographic look, sharp focus on the face. Vertical 9:16 frame.
```

**소재는 한 장만 씁니다.** 여러 장을 넣으면 개가 둘로 나옵니다.

마음에 안 들면 이 단계에서 다시 뽑으십시오. 소재가 흔들리면 아래 컷도 전부 흔들립니다.

---

## 3. 음성 준비

| 화자 | 톤 | 설정 |
| --- | --- | --- |
| 망고 | 밝고 또렷한 여자아이 | 커스텀 음성 저장 |
| 리포터 | 차분한 중년 남성 | 프롬프트 문장으로만 |

### 망고 음성 만들기

모델명 → Omni Flash → Add → Voices → Create New Voice

- Base Voice: 밝은 여성 음성 중 하나
- Name: `Mango`
- Voice Performance: `Make the voice sound like a cheerful young girl, bright and clear, slightly high-pitched.`
- Sample Dialogue: `가을날씨 너무 좋아서 놀러왔어요`

저장하면 `@Voice: Mango`로 부릅니다.

**리포터는 음성 참조를 걸지 않습니다.** Shot 1에만 나오고 프롬프트 문장으로 톤을 정합니다.

---

## 4. Shot 1 / 4초 — 목소리만 들린다

### 라벨 스캐폴드

```
[Cinematography]   Medium close-up, eye-level, static camera, shallow depth of
field.
[Subject]          @Mango, a small white Maltese dog with long straight white ear
fur hanging past its jaw, a short-clipped white body coat, large round dark brown
eyes and a dark brown nose, wearing a pink knit top with a white heart pattern and
ruffled shoulder trim, the only dog in the shot.
[Action]           The dog holds still and looks into the lens, ears twitching as
an unseen man speaks off-screen.
[Context]          An amusement park entrance plaza on a clear autumn afternoon,
blurred crowd far behind.
[Style & Ambiance] Handheld street-interview look, natural daylight, warm autumn
color, realistic.
Dialogue:          A calm middle-aged male reporter's voice, off-screen, says,
"여기가 어디에요?"
SFX / Ambient:     distant park crowd chatter, light autumn breeze.
Negative:          no microphone in frame, no person visible, no on-screen text,
exactly one dog in the scene, no restyling of the dog's face or top.
```

### Flow 입력용 (산문)

```
Medium close-up, eye-level, static camera with shallow depth of field. @Mango, a
small white Maltese dog with long straight white ear fur hanging past its jaw, a
short-clipped white body coat, large round dark brown eyes and a dark brown nose,
wearing a pink knit top with a white heart pattern and ruffled shoulder trim,
faces the camera straight on and holds still, ears twitching as an unseen man
speaks off-screen. An amusement park entrance plaza on a clear autumn afternoon, a
blurred crowd and ride structures far behind. Handheld street-interview look,
natural daylight, warm autumn color, realistic. A calm middle-aged male reporter's
voice, off-screen and never shown, says, "여기가 어디에요?" Ambient noise: distant park
crowd chatter and a light autumn breeze. Keep @Mango's face, ear fur and pink
heart-patterned top exactly as they are in the reference image and do not restyle,
recolor or replace them. Keep the frame clear of any microphone, any person and
any on-screen text or subtitles. Exactly one dog is in the scene and no second dog
appears anywhere, not even in the background.
```

**음성 참조를 걸지 않습니다.** 리포터 목소리이기 때문입니다.

---

## 5. Shot 2 / 6초 — 마이크가 들어온다

### 라벨 스캐폴드

```
[Cinematography]   Medium close-up, eye-level, static camera.
[Subject]          @Mango, a small white Maltese dog with long straight white ear
fur hanging past its jaw, a short-clipped white body coat, large round dark brown
eyes and a dark brown nose, wearing a pink knit top with a white heart pattern and
ruffled shoulder trim, the only dog in the shot.
[Action]           A black foam-topped handheld microphone rises into the bottom
of the frame; the dog lifts its chin and speaks clearly into it.
[Context]          The same amusement park entrance plaza, autumn afternoon.
[Style & Ambiance] Street-interview realism, natural daylight, warm tone.
Dialogue:          @Voice: Mango. A bright, clear young girl's voice says, "가을날씨
너무 좋아서 에버랜드 놀러왔어요"
SFX / Ambient:     soft microphone handling noise, distant park crowd.
Negative:          only the microphone enters the frame, the reporter stays out of
shot, no on-screen text, exactly one dog in the scene.
```

### Flow 입력용 (산문)

```
Medium close-up, eye-level, static camera. @Mango, a small white Maltese dog with
long straight white ear fur hanging past its jaw, a short-clipped white body coat,
large round dark brown eyes and a dark brown nose, wearing a pink knit top with a
white heart pattern and ruffled shoulder trim, waits in frame as a black foam-
topped handheld microphone rises into the bottom of the shot and moves toward the
dog's muzzle. The dog lifts its chin slightly and speaks clearly into it. The same
amusement park entrance plaza, autumn afternoon, street-interview realism, natural
daylight, warm tone. @Voice: Mango. A bright, clear young girl's voice says, "가을날씨
너무 좋아서 에버랜드 놀러왔어요" SFX: soft microphone handling noise over distant park crowd.
Keep @Mango's face, ear fur and pink heart-patterned top exactly as they are in
the reference image and do not restyle, recolor or replace them. Only the
microphone enters the frame, the reporter stays out of shot, and there is no on-
screen text. Exactly one dog is in the scene and no second dog appears anywhere,
not even in the background.
```

**왜 6초인가**

대사가 21자입니다. 4초에는 안 들어갑니다.

6초로 만들고 나중에 자릅니다.

짧게 가려면 `가을날씨 좋아서 에버랜드 왔어요`로 줄이면 4초에 맞습니다.

---

## 6. Shot 3 / 4초 — 실제 탑승 영상

### 라벨 스캐폴드

```
[Cinematography]   Tight front-facing on-ride camera shot, eye-level with the dog,
heavy camera shake.
[Subject]          @Mango, a small white Maltese dog with long straight white ear
fur hanging past its jaw, a short-clipped white body coat, large round dark brown
eyes and a dark brown nose, wearing a pink knit top with a white heart pattern and
ruffled shoulder trim, the only dog in the shot.
[Action]           The coaster car plunges down a steep drop; the dog's ear fur
blows straight up and its mouth opens wide in a scream.
[Context]          A large wooden roller coaster mid-descent, sky and park scenery
rushing past.
[Style & Ambiance] Raw on-ride footage look, harsh daylight, motion blur, slightly
grainy.
Dialogue:          @Voice: Mango. A bright young girl's voice screams, "으아아아 살려줘
멍!!"
SFX / Ambient:     roaring wind, rattling coaster track.
Negative:          no other passengers in frame, no on-screen text, exactly one
dog in the scene.
```

### Flow 입력용 (산문)

```
Tight front-facing on-ride camera shot, eye-level with the dog, heavy camera
shake. @Mango, a small white Maltese dog with long straight white ear fur hanging
past its jaw, a short-clipped white body coat, large round dark brown eyes and a
dark brown nose, wearing a pink knit top with a white heart pattern and ruffled
shoulder trim, is strapped into a roller coaster seat as the car plunges down a
steep drop; its ear fur blows straight up and its mouth opens wide in a scream. A
large wooden roller coaster mid-descent, sky and park scenery rushing past. Raw
on-ride footage look, harsh daylight, motion blur, slightly grainy. @Voice: Mango.
A bright young girl's voice screams, "으아아아 살려줘 멍!!" SFX: roaring wind and rattling
coaster track. Keep @Mango's face, ear fur and pink heart-patterned top exactly as
they are in the reference image and do not restyle, recolor or replace them. No
other passengers in frame and no on-screen text. Exactly one dog is in the scene
and no second dog appears anywhere, not even in the background.
```

---

## 7. Shot 4 / 4초 — 시치미

### 라벨 스캐폴드

```
[Cinematography]   Medium close-up, eye-level, static camera, shallow depth of
field.
[Subject]          @Mango, a small white Maltese dog with long straight white ear
fur hanging past its jaw, a short-clipped white body coat, large round dark brown
eyes and a dark brown nose, wearing a pink knit top with a white heart pattern and
ruffled shoulder trim, the only dog in the shot.
[Action]           The dog smooths its ear fur back with one front paw and puts on
a composed face, but the paw keeps trembling.
[Context]          The same park plaza, autumn afternoon.
[Style & Ambiance] Street-interview realism, natural daylight, warm tone.
Dialogue:          @Voice: Mango. A bright young girl's voice says, "그건 편집으로 좀
잘라라."
SFX / Ambient:     distant park crowd.
Negative:          no on-screen text, no person in frame, exactly one dog in the
scene.
```

### Flow 입력용 (산문)

```
Medium close-up, eye-level, static camera with shallow depth of field. @Mango, a
small white Maltese dog with long straight white ear fur hanging past its jaw, a
short-clipped white body coat, large round dark brown eyes and a dark brown nose,
wearing a pink knit top with a white heart pattern and ruffled shoulder trim, is
back in the interview spot. The dog smooths its ear fur back with one front paw
and puts on a composed face, but the paw keeps trembling. The same park plaza,
autumn afternoon, street-interview realism, natural daylight, warm tone. @Voice:
Mango. A bright young girl's voice says, "그건 편집으로 좀 잘라라." SFX: distant park crowd.
Keep @Mango's face, ear fur and pink heart-patterned top exactly as they are in
the reference image and do not restyle, recolor or replace them. No on-screen text
and no person in frame. Exactly one dog is in the scene and no second dog appears
anywhere, not even in the background.
```

**동작을 바꿨습니다**

원래는 선글라스를 고쳐 쓰는 동작이었습니다.

선글라스가 없으니 **앞발로 귀 털을 쓸어넘기는 동작**으로 바꿨습니다.

시치미 떼는 느낌은 그대로입니다.

**대사도 줄였습니다**

원래 `그건 편집으로 좀 잘라라. 다시 갑시다 멍.`이었습니다. 4초에 너무 깁니다.

---

## 8. Shot 5 / 6초 — 집으로

### 라벨 스캐폴드

```
[Cinematography]   Close-up, slightly high angle looking down at the back seat,
static camera.
[Subject]          @Mango, a small white Maltese dog with long straight white ear
fur hanging past its jaw, a short-clipped white body coat, large round dark brown
eyes and a dark brown nose, wearing a pink knit top with a white heart pattern and
ruffled shoulder trim, the only dog in the shot.
[Action]           The dog is sprawled flat asleep on the back seat, chest rising
slowly, mumbling in its sleep.
[Context]          The back seat of a moving car at dusk, street lights sliding
past the window.
[Style & Ambiance] Warm dim evening light, cozy and soft, shallow depth of field,
realistic.
Dialogue:          @Voice: Mango. A bright young girl's voice, sleepy and soft,
says, "다음엔 어디 털러 갈까?"
SFX / Ambient:     low road hum.
Negative:          no driver visible, no on-screen text, exactly one dog in the
scene.
```

### Flow 입력용 (산문)

```
Close-up from a slightly high angle looking down at the back seat, static camera.
@Mango, a small white Maltese dog with long straight white ear fur hanging past
its jaw, a short-clipped white body coat, large round dark brown eyes and a dark
brown nose, wearing a pink knit top with a white heart pattern and ruffled
shoulder trim, is sprawled flat asleep on the car's back seat, chest rising
slowly, mumbling in its sleep. The back seat of a moving car at dusk, street
lights sliding past the window. Warm dim evening light, cozy and soft, shallow
depth of field, realistic. @Voice: Mango. A bright young girl's voice, sleepy and
soft, says, "다음엔 어디 털러 갈까?" SFX: low road hum. Keep @Mango's face, ear fur and
pink heart-patterned top exactly as they are in the reference image and do not
restyle, recolor or replace them. No driver visible and no on-screen text. Exactly
one dog is in the scene and no second dog appears anywhere, not even in the
background.
```

**뒷부분은 자막으로 넣습니다**

`팔로우하고 기다려라 멍!`과 CTA 자막은 편집 앱에서 올립니다.

Flow가 만드는 글자는 자주 깨집니다. 한글은 특히 그렇습니다.

---

## 9. 이어 붙이기

Scenebuilder에서 순서대로 놓고 앞뒤를 잘라냅니다.

| 샷 | 생성 | 최종 |
| --- | --- | --- |
| 1 | 4초 | 2초 |
| 2 | 6초 | 2초 |
| 3 | 4초 | 3초 |
| 4 | 4초 | 3초 |
| 5 | 6초 | 2초 |
| 합계 | 24초 | 12초 |

클립을 넣으려면 클립에 마우스를 올리고 More → Add to Scene을 고릅니다.

**루프 연결**

Shot 5의 잠든 얼굴을 Shot 1의 클로즈업에 붙입니다.

꿈에서 깨는 것처럼 이어집니다.

---

## 10. 그래도 흔들리면

조명이 크게 다른 컷이 가장 많이 흔들립니다. Shot 3(탑승)과 Shot 5(해질녘 차)입니다.

그 컷만 이렇게 해보십시오.

1. Nano Banana Pro로 그 장면의 정지 이미지를 먼저 만듭니다.
2. 소재 이미지를 올리고 "같은 개"를 지시합니다.
3. 나온 이미지를 시작 프레임(Frames)으로 넣고 영상을 만듭니다.

**단, 시작 프레임만 쓰면 `@Voice`를 못 씁니다.**

대사가 있는 컷은 소재 방식을 유지하십시오.

---

## 11. 알아둘 것

### 워터마크

한국 거주자에게는 가시 워터마크가 자동으로 붙습니다. 끌 수 없습니다.

모든 출력에 비가시 SynthID도 들어갑니다.

### 립싱크가 흔들리면

1. 대사를 더 줄입니다.
2. 클립 길이를 6초나 8초로 늘립니다.
3. 그래도 안 되면 샷을 둘로 나눕니다.

### 가정한 값

| 항목 | 값 | 근거 |
| --- | --- | --- |
| 화면비 | 9:16 | 인스타 릴스 |
| 시간대 | 가을 오후 | 대사 내용 |
| 마지막 컷 | 해질녘 | 집에 가는 길 |
| 옷 | 분홍 하트 | 2023년 사진 |

이 값들은 직접 고쳐 쓰셔도 됩니다.