# Civitai 무료 생성 프롬프트 키트 (v1.2 · 2026-10-02)

Civitai 무료 계정으로 바로 붙여 쓰는 프롬프트입니다. GPT 이미지 1장으로 3샷이 들어간 H3 9초 영상을 한 번에 만드는 구성과, 항목마다 실제 생성한 비용·시간·결과를 적었습니다.

## 1. NARI 9초 3샷 공식

- 모델: MiniMax H3 (Comfy) · 표준 30 step · 9초
- 용도: 한 번 생성으로 훅·환기·CTA 3초씩 나눈 숏폼
- 실측: 대괄호만 바꿔 씁니다. 샷마다 구도·행동 하나·대사 한 줄(3초 분량), 의상은 문장으로 고정.

```text
Vertical 9:16 UGC short, three shots joined by hard cuts.
Keep <Subject 1> consistent with <Picture 1> in every shot. Outfit, exactly as in <Picture 1>: [outfit in words]. Do not change it.
[Props/hands rule, e.g. right hand keeps the microphone near her chin; only the other hand moves.]
Spoken [language] exists only as audio. No visible text or captions.

<Shot 1> 00:00.00-00:03.00 (hook)
[framing], eye-level, locked camera. [one action on frame one].
Dialogue for this shot, spoken aloud only: <d>[[language]] [hook line, ~3 s]</d>

<Shot 2> 00:03.00-00:06.00 (reset)
A clean hard cut to [a different framing]. [one action].
Dialogue for this shot, spoken aloud only: <d>[[language]] [middle line, ~3 s]</d>

<Shot 3> 00:06.00-00:09.00 (call to action)
A clean hard cut to [another framing]. [one action, e.g. points down once].
Dialogue for this shot, spoken aloud only: <d>[[language]] [CTA line, ~3 s]</d>
```

## 2. NARI 9초 3샷 (한 번에 생성)

- 모델: MiniMax H3 (Comfy) · 표준 30 step · 9초
- 용도: GPT 레퍼런스 1장 → 9초 숏폼
- 실측: 315 blue · 433.5초 · 9.42초 출력. 하드컷 3.29·6.88초, 대사 3줄 일치, 의상 유지. 의상 문장을 뺀 1차는 카디건 색이 바뀜. 컷·무음 다듬어 8.67초로 사용.

```text
subject_definitions:
<Subject 1> is the woman in <Picture 1>.
picture_definitions:
<Picture 1> is the reference image for <Subject 1>, her outfit, the handheld microphone, the chair and the room.
summary:
A 9-second vertical 9:16 UGC short made of three clearly separated 3-second shots joined by hard cuts: a hook, a reset and a call to action.
Keep <Subject 1> consistent with <Picture 1> in every shot: same face, hairstyle, outfit, microphone, chair and room.
Outfit, exactly as in <Picture 1> in every shot: a cream cable-knit cardigan over a white T-shirt and light blue jeans. Do not change the cardigan color.
Her right hand keeps holding the small black microphone near her chin for the entire video. Only her other hand may move.
Spoken Korean exists only as audio. No visible text, subtitles, captions or logos.

detailed_description:

<Shot 1> 00:00.00-00:03.00 (hook)
Medium-close seated shot, eye-level, locked camera.
On frame one she leans slightly toward the lens with raised eyebrows and starts speaking immediately, surprised and fast.
Dialogue for this shot, spoken aloud only: <d>[Korean] 이 영상, 돈 한 푼 안 썼어요.</d>

<Shot 2> 00:03.00-00:06.00 (reset)
A clean hard cut to a closer shot of her face and upper torso, eye-level, locked camera.
She has just glanced at her monitor and turns back to the lens, calm and friendly, her free hand making one small open-palm gesture.
Dialogue for this shot, spoken aloud only: <d>[Korean] 매일 받는 무료 포인트로 만들었죠.</d>

<Shot 3> 00:06.00-00:09.00 (call to action)
A clean hard cut back to a medium seated shot, eye-level, locked camera.
She smiles warmly at the lens and points her free index finger downward once, bright and inviting, then gives a small nod.
Dialogue for this shot, spoken aloud only: <d>[Korean] 방법은 프로필 링크에 있어요.</d>

visual_style:
Vertical 9:16. UGC creator style. Clear three-shot structure with two hard cuts at 3 and 6 seconds.
Plain realistic creator-camera footage; the frame contains only the person and the room, exactly like <Picture 1>.
```

## 3. ChatGPT 레퍼런스 이미지 요청 (형식 예시)

- 모델: ChatGPT 이미지(무료 플랜)
- 용도: 9초 영상에 쓰는 인물 레퍼런스 1장
- 실측: NARI 레퍼런스(941×1672)는 이 구성(인물·의상·소품·방·조명)으로 ChatGPT 이미지에서 만들었습니다. 이 문구 자체를 그대로 다시 생성해 본 것은 아닙니다.

```text
Vertical 9:16 photo of a young Korean woman, a short-form video creator, sitting in a cozy beige armchair in her home studio and looking at the camera.
She holds a small black handheld microphone near her chin with her right hand.
Outfit: cream cable-knit cardigan over a white T-shirt and light blue jeans.
Room: warm lamps, plants, a wall of photos and a monitor showing an editing timeline behind her.
Realistic smartphone photo, eye-level, full upper body and knees in frame, soft warm evening light. No text or logos.
```

## 4. 이미지 프롬프트 공식

- 모델: Qwen 2.1
- 용도: 모든 이미지의 뼈대. 아래 예시 5개가 이 순서를 따릅니다.
- 실측: 순서만 지키면 짧아도 됩니다. 한 문장에 한 요소씩, 쉼표로 나눕니다.

```text
[Subject: who or what, with 2-3 visible details],
[Action or pose],
[Place and background],
[Light: direction and mood],
[Camera: framing, angle, lens],
[Style and texture]
```

## 5. 인물 · 세로 프로필 사진

- 모델: Qwen 2.1
- 용도: 크리에이터 프로필, 영상 첫 프레임
- 실측: 5 blue · 23.7초. 손·머그·노트북 형태 정상. 이 이미지를 아래 영상 예시의 레퍼런스로 사용.

```text
A friendly young woman in her late 20s with shoulder-length dark hair, wearing a cream knit sweater, sitting at a wooden desk and smiling at the camera, holding a coffee mug. Cozy home office with a laptop and a small plant, soft morning window light from the left, shallow depth of field. Vertical smartphone photo, eye-level, medium close-up, natural skin texture, realistic photography.
```

## 6. 제품 사진

- 모델: Qwen 2.1
- 용도: 상세 페이지·썸네일용 제품 컷
- 실측: 5 blue · 23.8초. 김·그림자 표현 정상. "matte"를 넣어도 약간의 광택은 남음.

```text
A matte white ceramic coffee cup on a light oak table, steam rising from the cup, a folded linen napkin beside it. Clean minimal background in warm beige, soft diffused studio light from above-left, gentle shadow. Product photography, 50mm lens, centered composition, high detail.
```

## 7. 한글 문구 포스터

- 모델: Qwen 2.1
- 용도: 썸네일·공지 이미지
- 실측: 5 blue · 29.6초. "100% 무료 AI", "이미지·영상 생성" 글자 그대로 정확히 렌더링.

```text
미니멀한 세로 포스터. 상단에 크고 굵은 한글 제목 "100% 무료 AI", 그 아래 작은 글씨 "이미지·영상 생성". 짙은 남색 배경에 부드러운 그라데이션 빛, 중앙에 빛나는 카메라 아이콘. 깔끔한 산세리프 타이포그래피, 플랫 벡터 스타일.
```

## 8. 영문 문구 포스터

- 모델: Qwen 2.1
- 용도: 영문 썸네일
- 실측: 5 blue · 57.0초. 배경 지시 없이 쓴 2회는 상단에 흰 띠가 생기고 제목이 잘리거나 숫자가 깨짐("420"→"4'0"). "full-bleed … no white border or band"를 넣은 이 버전에서 해결.

```text
A minimalist vertical poster on a full-bleed deep navy background that fills the entire frame edge to edge, no white border or band. Large bold headline text "100% FREE AI" near the top, with the smaller line "images and video, every day" below it. Center illustration of a glowing camera icon with soft gradient light. Clean sans-serif typography, flat vector style.
```

## 9. 사람 없는 배경

- 모델: Qwen 2.1
- 용도: 영상 배경, 합성용 배경판
- 실측: 5 blue · 46.1초(대기열 포함). "no people"로 인물 없음 확인.

```text
An empty cozy cafe interior at golden hour, wooden tables, hanging warm pendant lights, large windows with a city street outside, no people. Wide establishing shot, eye-level, warm color palette, cinematic lighting, realistic photography, vertical frame.
```

## 10. 영상 프롬프트 공식 (레퍼런스 방식)

- 모델: MiniMax H3 (Comfy) · Turbo
- 용도: 말하는 인물 4초 영상의 뼈대
- 실측: 4초에 대사 8~11단어가 들어갑니다. 대사는 큰따옴표로 감쌉니다.

```text
Use <Picture 1> as the only visual reference for [subject], [outfit], [props] and [room].
Preserve the exact framing and subject size from <Picture 1>. Completely static locked camera: no zoom, no crop change, no cut.
[Subject] looks into the camera and says in a [tone] voice: "[one short sentence, 8-11 words for 4 seconds]." [One small closing action].
Motion: subtle and natural only. [Hands/props] stay still. No fast gestures.
```

## 11. 말하는 인물 4초 (레퍼런스 방식, 추천)

- 모델: MiniMax H3 (Comfy) · Turbo
- 용도: 이미지 1장 → 대사가 있는 숏폼 컷
- 실측: 48 blue · 52.6초 · 4.46초 출력(소리 포함). 대사 10단어 전부 일치, 컷 없이 안정. 단, 원본보다 조금 가까운 구도로 다시 잡힘.

```text
Use <Picture 1> as the only visual reference for the woman, her outfit, the mug and the room.
Preserve the exact framing and subject size from <Picture 1>. Completely static locked camera: no zoom, no crop change, no cut.
She looks into the camera and says in a warm, clear voice: "I make all of this for free, every single day." She smiles and gives a small nod at the end.
Motion: subtle and natural only. Both hands stay still around the mug. No fast gestures.
```

## 12. 말하는 인물 4초 (첫 프레임 방식)

- 모델: MiniMax H3 (Comfy) · Turbo
- 용도: 원본 이미지 그대로 시작해야 할 때
- 실측: 48 blue · 42.1초. 대사 일치, 손 번짐 없음. 그러나 0.5초 안에 확대 컷이 생김 → "static"을 써도 막히지 않음.

```text
The woman in the image looks into the camera and says in a warm, clear voice: "I make all of this for free, every single day." She smiles and gives a small nod at the end.
Camera: completely static, locked tripod shot, no zoom or reframing.
Motion: subtle and natural only. Both hands stay still around the mug. No fast gestures.
Keep her face, hair, outfit and the room exactly as in the image.
```

https://resary.com/ko/guides/civitai-free-daily

---

# Civitai Free Generation Prompt Kit (v1.2 · 2026-10-02)

Prompts you can paste into a free Civitai account: one GPT image turned into a three-shot, 9-second H3 short in a single generation, with the cost, time and result we got for every item.

## 1. NARI 9-second 3-shot formula

- Model: MiniMax H3 (Comfy) · standard 30 steps · 9 s
- Use: One generation split into hook, reset and CTA at 3 s each
- Tested: Replace only the brackets. One framing, one action and one ~3-second line per shot; pin the outfit in words.

```text
Vertical 9:16 UGC short, three shots joined by hard cuts.
Keep <Subject 1> consistent with <Picture 1> in every shot. Outfit, exactly as in <Picture 1>: [outfit in words]. Do not change it.
[Props/hands rule, e.g. right hand keeps the microphone near her chin; only the other hand moves.]
Spoken [language] exists only as audio. No visible text or captions.

<Shot 1> 00:00.00-00:03.00 (hook)
[framing], eye-level, locked camera. [one action on frame one].
Dialogue for this shot, spoken aloud only: <d>[[language]] [hook line, ~3 s]</d>

<Shot 2> 00:03.00-00:06.00 (reset)
A clean hard cut to [a different framing]. [one action].
Dialogue for this shot, spoken aloud only: <d>[[language]] [middle line, ~3 s]</d>

<Shot 3> 00:06.00-00:09.00 (call to action)
A clean hard cut to [another framing]. [one action, e.g. points down once].
Dialogue for this shot, spoken aloud only: <d>[[language]] [CTA line, ~3 s]</d>
```

## 2. NARI 9-second 3-shot (one generation)

- Model: MiniMax H3 (Comfy) · standard 30 steps · 9 s
- Use: One GPT reference → a 9-second short
- Tested: 315 blue · 433.5 s · 9.42 s output. Hard cuts at 3.29 and 6.88 s, all 3 lines matched, outfit kept. Run 1 without the outfit sentence changed the cardigan color. Trimmed to 8.67 s for use.

```text
subject_definitions:
<Subject 1> is the woman in <Picture 1>.
picture_definitions:
<Picture 1> is the reference image for <Subject 1>, her outfit, the handheld microphone, the chair and the room.
summary:
A 9-second vertical 9:16 UGC short made of three clearly separated 3-second shots joined by hard cuts: a hook, a reset and a call to action.
Keep <Subject 1> consistent with <Picture 1> in every shot: same face, hairstyle, outfit, microphone, chair and room.
Outfit, exactly as in <Picture 1> in every shot: a cream cable-knit cardigan over a white T-shirt and light blue jeans. Do not change the cardigan color.
Her right hand keeps holding the small black microphone near her chin for the entire video. Only her other hand may move.
Spoken Korean exists only as audio. No visible text, subtitles, captions or logos.

detailed_description:

<Shot 1> 00:00.00-00:03.00 (hook)
Medium-close seated shot, eye-level, locked camera.
On frame one she leans slightly toward the lens with raised eyebrows and starts speaking immediately, surprised and fast.
Dialogue for this shot, spoken aloud only: <d>[Korean] 이 영상, 돈 한 푼 안 썼어요.</d>

<Shot 2> 00:03.00-00:06.00 (reset)
A clean hard cut to a closer shot of her face and upper torso, eye-level, locked camera.
She has just glanced at her monitor and turns back to the lens, calm and friendly, her free hand making one small open-palm gesture.
Dialogue for this shot, spoken aloud only: <d>[Korean] 매일 받는 무료 포인트로 만들었죠.</d>

<Shot 3> 00:06.00-00:09.00 (call to action)
A clean hard cut back to a medium seated shot, eye-level, locked camera.
She smiles warmly at the lens and points her free index finger downward once, bright and inviting, then gives a small nod.
Dialogue for this shot, spoken aloud only: <d>[Korean] 방법은 프로필 링크에 있어요.</d>

visual_style:
Vertical 9:16. UGC creator style. Clear three-shot structure with two hard cuts at 3 and 6 seconds.
Plain realistic creator-camera footage; the frame contains only the person and the room, exactly like <Picture 1>.
```

## 3. ChatGPT reference image request (format example)

- Model: ChatGPT image (free plan)
- Use: The one character reference for the 9-second video
- Tested: The NARI reference (941×1672) was made in ChatGPT image with this structure (person, outfit, prop, room, light). We did not regenerate this exact wording.

```text
Vertical 9:16 photo of a young Korean woman, a short-form video creator, sitting in a cozy beige armchair in her home studio and looking at the camera.
She holds a small black handheld microphone near her chin with her right hand.
Outfit: cream cable-knit cardigan over a white T-shirt and light blue jeans.
Room: warm lamps, plants, a wall of photos and a monitor showing an editing timeline behind her.
Realistic smartphone photo, eye-level, full upper body and knees in frame, soft warm evening light. No text or logos.
```

## 4. Image prompt formula

- Model: Qwen 2.1
- Use: The skeleton for every image. All five examples below follow this order.
- Tested: Short is fine as long as the order holds. One element per clause, separated by commas.

```text
[Subject: who or what, with 2-3 visible details],
[Action or pose],
[Place and background],
[Light: direction and mood],
[Camera: framing, angle, lens],
[Style and texture]
```

## 5. Portrait · vertical profile photo

- Model: Qwen 2.1
- Use: Creator profile, first frame for video
- Tested: 5 blue · 23.7 s. Hands, mug and laptop render correctly. Used as the reference for the video examples below.

```text
A friendly young woman in her late 20s with shoulder-length dark hair, wearing a cream knit sweater, sitting at a wooden desk and smiling at the camera, holding a coffee mug. Cozy home office with a laptop and a small plant, soft morning window light from the left, shallow depth of field. Vertical smartphone photo, eye-level, medium close-up, natural skin texture, realistic photography.
```

## 6. Product photo

- Model: Qwen 2.1
- Use: Product shots for pages and thumbnails
- Tested: 5 blue · 23.8 s. Steam and shadow look right. "matte" still leaves a slight sheen.

```text
A matte white ceramic coffee cup on a light oak table, steam rising from the cup, a folded linen napkin beside it. Clean minimal background in warm beige, soft diffused studio light from above-left, gentle shadow. Product photography, 50mm lens, centered composition, high detail.
```

## 7. Korean text poster

- Model: Qwen 2.1
- Use: Thumbnails and announcements
- Tested: 5 blue · 29.6 s. Both Korean lines rendered exactly as written.

```text
미니멀한 세로 포스터. 상단에 크고 굵은 한글 제목 "100% 무료 AI", 그 아래 작은 글씨 "이미지·영상 생성". 짙은 남색 배경에 부드러운 그라데이션 빛, 중앙에 빛나는 카메라 아이콘. 깔끔한 산세리프 타이포그래피, 플랫 벡터 스타일.
```

## 8. English text poster

- Model: Qwen 2.1
- Use: English thumbnails
- Tested: 5 blue · 57.0 s. Two earlier runs without the background instruction got a white band at the top, a clipped headline or a broken number ("420"→"4'0"). Adding "full-bleed … no white border or band" fixed it in this version.

```text
A minimalist vertical poster on a full-bleed deep navy background that fills the entire frame edge to edge, no white border or band. Large bold headline text "100% FREE AI" near the top, with the smaller line "images and video, every day" below it. Center illustration of a glowing camera icon with soft gradient light. Clean sans-serif typography, flat vector style.
```

## 9. Empty background

- Model: Qwen 2.1
- Use: Video backgrounds and compositing plates
- Tested: 5 blue · 46.1 s (including queue). "no people" kept the scene empty.

```text
An empty cozy cafe interior at golden hour, wooden tables, hanging warm pendant lights, large windows with a city street outside, no people. Wide establishing shot, eye-level, warm color palette, cinematic lighting, realistic photography, vertical frame.
```

## 10. Video prompt formula (reference mode)

- Model: MiniMax H3 (Comfy) · Turbo
- Use: Skeleton for a 4-second talking clip
- Tested: 4 seconds fits 8–11 spoken words. Put the line in double quotes.

```text
Use <Picture 1> as the only visual reference for [subject], [outfit], [props] and [room].
Preserve the exact framing and subject size from <Picture 1>. Completely static locked camera: no zoom, no crop change, no cut.
[Subject] looks into the camera and says in a [tone] voice: "[one short sentence, 8-11 words for 4 seconds]." [One small closing action].
Motion: subtle and natural only. [Hands/props] stay still. No fast gestures.
```

## 11. Talking clip, 4 s (reference mode, recommended)

- Model: MiniMax H3 (Comfy) · Turbo
- Use: One image → a short-form clip with dialogue
- Tested: 48 blue · 52.6 s · 4.46 s output with audio. All 10 words matched, no cuts. The shot is re-framed slightly tighter than the source.

```text
Use <Picture 1> as the only visual reference for the woman, her outfit, the mug and the room.
Preserve the exact framing and subject size from <Picture 1>. Completely static locked camera: no zoom, no crop change, no cut.
She looks into the camera and says in a warm, clear voice: "I make all of this for free, every single day." She smiles and gives a small nod at the end.
Motion: subtle and natural only. Both hands stay still around the mug. No fast gestures.
```

## 12. Talking clip, 4 s (first-frame mode)

- Model: MiniMax H3 (Comfy) · Turbo
- Use: When the clip must start on the exact image
- Tested: 48 blue · 42.1 s. Dialogue matched, no hand smearing. But it cut to a closer shot within 0.5 s — "static" did not prevent it.

```text
The woman in the image looks into the camera and says in a warm, clear voice: "I make all of this for free, every single day." She smiles and gives a small nod at the end.
Camera: completely static, locked tripod shot, no zoom or reframing.
Motion: subtle and natural only. Both hands stay still around the mug. No fast gestures.
Keep her face, hair, outfit and the room exactly as in the image.
```

https://resary.com/en/guides/civitai-free-daily