龙虾大学skill
龙虾电影分镜海报 Skill
# 龙虾电影分镜海报 Skill
> 人物一致性 × 动漫镜头语言 × 流光溢彩 × 品牌叙事。
## 快速说明
# 龙虾电影分镜海报（lobster-cinematic-keyframe-poster）
人物一致性 × 动漫镜头语言 × 流光溢彩 × 品牌叙事。
## 适合做什么
- 把人物参考图做成电影关键帧海报。
- 做面部/眼睛特写、回眸、动作大片、多分镜同场。
- 做品牌人物海报，例如酒业、课程、个人 IP、AI Energy Hub 系列视觉。
- 避免 AI 生图常见问题：正面证件照、星星噪点、中文乱字、脸不像。
## 核心方法
1. 参考图只当身份锚点。
2. 用 action pose 和 camera language 防止正面僵硬。
3. 用 split-panel / triptych / floating frames 做分镜感。
4. 用 iridescent light ribbons、amber liquid glow、caustic reflections 替代 dense starfield。
5. 中文文案短句独立声明；如果必须准确，生成无字底图后期排版。
## 触发示例
- “这张太像正面照了，做成电影海报，有动作。”
- “做几张眼睛特写、面部特写，像动漫关键帧那种。”
- “这个品牌人物海报要流光溢彩，但不要很多星星。”
- “同一个人做 5 张不同分镜的大片。”
## SKILL.md
```markdown
---
name: lobster-cinematic-keyframe-poster
description: This skill should be used when creating cinematic keyframe character posters, anime-inspired storyboard-style brand visuals, close-up face and eye posters, dynamic heroine action posters, or image-to-image prompt systems that preserve identity while using multi-shot composition, luminous light-flow effects, and premium Chinese brand narrative.
agent_created: true
version: 1.0.0
---
# 龙虾电影分镜海报 Skill
## 核心定位
把普通人物参考图或定妆图，升级为「电影关键帧海报 / 动漫镜头语言海报 / 品牌叙事视觉」。
核心不是单纯把人变美，而是同时完成四件事：
1. **身份一致性**：参考图只作为 identity anchor，保留脸、眼睛、肤色、年龄感、气质。
2. **镜头叙事**：从静态正面照升级为动作、回眸、伸手触光、眼部特写、多分镜同场。
3. **流光溢彩**：用低噪点的流光、琥珀液体光、棱镜轮廓光、水晶反射、丝带光效替代满天星星。
4. **品牌承载**：让人物成为品牌精神的化身，产品和场景作为叙事符号，不抢主体。
## 适用场景
处理以下任务时使用本技能：
- 基于人物照片生成电影感人物海报、品牌海报、酒业/教育/课程/个人 IP 海报。
- 用户反馈「太像正面照」「动作不够」「画面不够丰富」「想要动漫感/分镜感/大片感」。
- 生成面部特写、眼睛特写、回眸特写、半脸侧光、三镜同场、split-panel / triptych 构图。
- 批量生成同一人物的 3-6 张系列海报，要求脸像、风格统一、场景不同。
- 需要把发光元素做高级，但避免星星过多、噪点过多、廉价 AI 闪粉感。
## 工作流
### 1. 先暂悬：判断这一张图真正缺什么
不要一上来堆关键词。先判断用户要解决的是哪类问题：
| 问题 | 本质 | 处理方式 |
|---|---|---|
| 全是正面照 | 缺动作叙事 | 使用 action pose + cinematic camera language |
| 脸不像 | 身份锚点弱 | 降低场景复杂度，强化 face/eyes identity lock |
| 画面不丰富 | 缺镜头层次 | 使用 foreground/midground/background 或 multi-shot panels |
| 星星太多像噪点 | 光效控制差 | 改用 iridescent ribbons / amber glow / caustic reflections |
| 中文文字乱码 | 文字任务过重 | 英文视觉提示 + 独立中文 text block，必要时后期排版 |
### 2. 固定身份锚点
每次图生图都加入：
```text
Use the reference image only as a strict identity anchor, not as a composition or background reference.
Preserve the same face, eyes, skin tone, hairstyle, age impression, body proportion, and recognizable personal identity.
Do not invent a different person. Do not copy the original background, lighting, pose, or low-quality photo texture.
```
如果用户特别要求「脸像」，再加：
```text
Identity priority is higher than costume, background, and visual effects.
Keep the face and eyes clearly recognizable, clean, naturally beautiful, and emotionally expressive.
```
### 3. 选择海报类型
按目标选择一个结构，不要混乱叠加。
#### A. 动态女主动作海报
用于解决「太正面、太僵硬」。
```text
Create a premium vertical 9:16 cinematic heroine action poster.
VERY IMPORTANT: avoid static front-facing portrait. Create a cinematic full-body or three-quarter body action scene with a clear readable face.
ACTION POSE: [walking through wind / reaching out to touch glowing jar / turning back on a bridge / holding lantern / swirling sleeves].
CAMERA: low-angle heroic shot, 35mm cinematic lens, dynamic diagonal composition, motion lines from wind and fabric, strong silhouette.
```
适合动作词：
- walking forward through wind
- reaching out to touch a glowing object
- turning back over the shoulder
- holding a lantern and leading the path
- swirling sleeves to open a banquet / portal / light field
- kneeling to inspect a glowing artifact
- stepping across a bridge / threshold
#### B. 面部 / 眼睛特写海报
用于解决「想要更帅、更美、更有眼神」。
```text
Create a premium vertical 9:16 cinematic close-up key visual poster.
Focus on the face and beautiful expressive eyes. Use a realistic cinematic face, not an anime face, but borrow anime-inspired camera language.
CAMERA: extreme close-up of eyes / half-face side light / over-the-shoulder turnback close-up / shallow depth of field / soft bokeh background.
The eyes should be handsome, beautiful, luminous, emotionally strong, and sharply focused.
```
#### C. 多分镜同场海报
用于解决「画面要像动漫一幕里有多个镜头」。
```text
Create a storyboard-like cinematic key visual with multiple shots inside one unified scene.
Use 2-3 integrated panels or floating cinematic frames, not a messy collage.
Main panel: face close-up with expressive eyes.
Secondary panel: action gesture in the same scene.
Background panel: wide establishing shot of the brand world.
Keep consistent lighting, color palette, and identity across all panels.
```
推荐构图：
- `split-panel close-up`: 左侧眼部特写，右侧动作场景。
- `triptych key visual`: 上方眼睛，中部半身动作，下方场景远景。
- `floating frames`: 透明玻璃片般的分镜悬浮在同一空间。
- `single-scene multi-exposure`: 同一场景中叠化出远景、中景、特写。
### 4. 使用低噪点流光系统
不要默认写 starry sky / glitter / many stars。改用更可控的发光元素：
```text
Use flowing iridescent light ribbons, amber liquid glow, prismatic rim light, soft jade and cyan highlights, wine-gold caustic reflections, glass-like reflections, luminous silk ribbons, subtle volumetric light, sparse luminous particles only.
No dense starfield, no glitter noise, no messy sparkles, no visual clutter around the face.
```
光效优先级：
1. 脸部柔光：让脸清楚、眼睛好看。
2. 轮廓光：让人物从背景中立起来。
3. 叙事光源：酒坛、灯、杯、门、河、文字区域发光。
4. 少量粒子：只做空间深度，不做噪点。
### 5. 品牌叙事写法
品牌海报不要让产品抢人物。用场景承载品牌精神：
```text
The character is the spiritual embodiment of the brand.
Brand elements appear as symbolic environment details, not as a hard advertisement.
Use [brand industry elements] as cultural symbols in the foreground and background: [wine cellar / ceramic jar / banquet table / amber wine / family craft / moonlit courtyard].
```
中文文案单独声明：
```text
TEXT TO RENDER:
Add only this exact Chinese copy: “眼中有光·杯里有山河”.
The Chinese text must be clear, readable, correctly written, elegant, and integrated into the poster.
Do not add any other Chinese or English text.
```
如果中文必须 100% 准确，先生成无字底图，再用本地排版或 PPT/设计工具叠字。
## 标准提示词模板
### 动态动作版
```text
Use the reference image only as a strict identity anchor, not as a composition or background reference.
Preserve the same face, eyes, skin tone, hairstyle, age impression, body proportion, and recognizable personal identity.
Do not invent a different person.
Create a premium vertical 9:16 cinematic heroine action poster for [BRAND/THEME].
VERY IMPORTANT: avoid static passport portrait or stiff front-facing standing pose.
ACTION POSE: [SPECIFIC_ACTION].
SCENE: [BRAND_WORLD_SCENE].
CAMERA: low-angle heroic shot, 35mm cinematic lens, dynamic diagonal composition, foreground/midground/background layering, wind and fabric motion, clear readable face.
LIGHTING: soft key light on the face, prismatic rim light, amber liquid glow, subtle volumetric light, flowing iridescent ribbons, sparse luminous particles only.
COLOR: [MAIN_COLOR] + [SECONDARY_COLOR] + warm gold highlights, premium unified palette.
STYLE: realistic cinematic photography, Chinese fantasy blockbuster, anime-inspired camera language but realistic face, luxury movie poster, high-end editorial retouching.
TEXT TO RENDER:
Add only this exact Chinese copy: “[CHINESE_COPY]”.
The text must be clear, readable, correctly written, elegant, and integrated into the poster.
Do not add any other text.
NEGATIVE: static passport portrait, stiff front-facing standing pose, changed identity, duplicated face, extra people, deformed hands, dense starfield, glitter noise, messy sparkles, clutter around the face, cheap cartoon style, random English letters, watermark, wrong Chinese characters.
```
### 特写分镜版
```text
Use the reference image only as a strict identity anchor, not as a composition or background reference.
Preserve the same face, eyes, skin tone, hairstyle, age impression, body proportion, and recognizable personal identity.
Identity priority is higher than costume, background, and visual effects.
Create a premium vertical 9:16 cinematic close-up key visual poster for [BRAND/THEME].
Focus on beautiful expressive eyes, handsome gaze, face close-up, and multiple storyboard-like shots inside one unified scene.
Use a realistic cinematic face, but borrow anime-inspired camera language.
COMPOSITION: [extreme eye close-up / half-face side light / triptych / split-panel / floating frames].
Main panel: sharp face or eye close-up with luminous emotional gaze.
Secondary panel: [ACTION_GESTURE] in the same scene.
Background panel: [WIDE_BRAND_WORLD] with soft depth of field.
Keep all panels visually unified, not a messy collage.
LIGHTFLOW: flowing iridescent light ribbons, amber liquid glow, wine-gold caustic reflections, prismatic rim light, glass-like reflections, soft jade/cyan highlights, sparse luminous particles only.
Avoid relying on stars. No dense starfield, no glitter noise, no messy sparkles.
STYLE: realistic cinematic photography, luxury Chinese fantasy movie key visual, anime-inspired storyboard composition, shallow depth of field, rich framing, clean premium poster design.
TEXT TO RENDER:
Add only this exact Chinese copy: “[CHINESE_COPY]”.
The text must be clear, readable, correctly written, elegant, and integrated into the poster.
Do not add any other text.
NEGATIVE: changed identity, anime face, doll face, stiff portrait, static ID photo, cluttered collage, dense starfield, glitter noise, too many sparkles, wrong Chinese characters, random letters, watermark.
```
## 质量自检清单
生成前检查：
- 是否明确 reference image only as strict identity anchor？
- 是否有具体动作，而不是只写 dynamic pose？
- 是否有 camera language：low-angle / close-up / split-panel / triptych / shallow depth？
- 是否保护 face and eyes？
- 是否用 lightflow 替代 dense stars？
- 是否写了 no static passport portrait？
- 中文文案是否短、独立、只出现一次？
- 品牌元素是否作为环境符号，而不是硬塞 logo/产品？
生成后判断：
- 第一眼是否是「电影海报」，而不是普通写真？
- 脸是否像？眼睛是否好看？
- 动作是否有叙事？
- 光效是否高级、干净、低噪点？
- 多分镜是否统一，而不是拼贴杂乱？
- 中文是否可读？如果不可读，改无字底图 + 后期排版。
## 推荐读取的参考文件
- `references/camera-language.md`：动作、镜头、构图词库。
- `references/lightflow-elements.md`：低噪点流光系统。
- `references/storyboard-compositions.md`：多分镜构图模板。
- `references/brand-narrative.md`：品牌叙事与中文文案写法。
## 推荐脚本
使用 `scripts/generate_image2_series.py` 作为 GPT-Image-2 图生图批量生成脚本模板。复制到项目交付目录后，填入 API Key、参考图、输出目录和 `CARDS` 列表再运行。不要把密钥写入 Skill 本体。
```
## Bundled Resources
### references/brand-narrative.md
```md
# Brand Narrative Reference
## Core principle
A brand poster should not feel like a hard advertisement. Let the character embody the brand spirit; let the product and industry elements become symbolic environment details.
## Brand role hierarchy
1. Character = spiritual embodiment / emotional anchor.
2. Scene = brand world and cultural field.
3. Product elements = symbolic details.
4. Copy = short, poetic, memorable.
## Wine brand example
Use wine elements as:
- amber liquid glow
- ceramic jars
- moonlit brewing courtyard
- cellar tunnel
- banquet table
- cup reflections
- family craft symbols
- mountain / river / hometown atmosphere
Avoid:
- giant product bottle blocking the face
- supermarket advertisement feeling
- too many labels or logos
- product packshot replacing narrative
## Copywriting patterns
Short Chinese copy works better for native text rendering.
Examples:
- 眼中有光·杯里有山河
- 半面流光·一酿成诗
- 一镜初心·三镜成章
- 回眸一瞬·万盏生光
- 月色入坛·匠心成酿
- 风起创基·一杯见山河
- 把星河酿进人间烟火
## Text prompt pattern
```text
TEXT TO RENDER:
Add only this exact Chinese copy: “[CHINESE_COPY]”.
The Chinese text must be clear, readable, correctly written, elegant, and integrated into the poster.
Do not add any other Chinese or English text.
```
## When text accuracy matters most
Generate a no-text poster first, then add Chinese copy with local layout tools. Image models may still distort Chinese, even with clear prompts.
```
### references/camera-language.md
```md
# Camera Language Reference
## Core principle
Use camera language to prevent static portrait outputs. A poster becomes cinematic when the prompt specifies action, lens, angle, body direction, and scene motivation.
## Anti-static phrases
```text
avoid static passport portrait
avoid stiff front-facing standing pose
not a studio ID photo
not a character sheet
not a simple beauty portrait
```
## Heroine action poses
- walking forward through wind, fabric and hair moving naturally
- turning back over the shoulder, eyes catching the key light
- reaching out to touch a glowing artifact
- holding a lantern while leading the viewer into the scene
- stepping across a bridge, threshold, portal, or moonlit courtyard
- swirling sleeves, opening light, banquet, or memory field
- kneeling beside an ancient jar, inspecting glowing liquid reflections
- lifting a cup / scroll / flower / crystal as a symbolic brand action
## Cinematic angles
- low-angle heroic shot
- three-quarter body shot
- over-the-shoulder turnback close-up
- extreme close-up of eyes
- half-face side-light portrait
- foreground object framing the face
- diagonal composition with motion direction
- shallow depth of field, face sharply focused
## Lens and composition vocabulary
- 35mm cinematic lens for action context
- 50mm portrait lens for natural face
- 85mm shallow depth portrait for face close-up
- anamorphic cinematic framing
- layered foreground / midground / background
- strong silhouette against luminous background
- controlled negative space for title text
- rich framing, but clean around the face
## Prompt snippets
```text
CAMERA: low-angle heroic shot, 35mm cinematic lens, dynamic diagonal composition, foreground/midground/background layering, wind and fabric motion, clear readable face.
```
```text
CAMERA: extreme close-up of her eyes, shallow depth of field, soft bokeh background, the eyes sharply focused with emotional luminous reflections.
```
```text
COMPOSITION: split-panel cinematic poster, main panel is a face close-up, secondary panel shows her hand touching a glowing artifact, background panel shows the wide brand world.
```
```
### references/lightflow-elements.md
```md
# Lightflow Elements Reference
## Core principle
Use controlled luminous materials instead of dense starfields. The goal is 流光溢彩, not visual noise.
## Preferred light elements
- flowing iridescent light ribbons
- amber liquid glow
- wine-gold caustic reflections
- prismatic rim light
- glass-like reflections
- luminous silk ribbons
- soft jade and cyan highlights
- subtle volumetric light
- glowing mist layers
- reflected light in eyes
- sparse luminous particles only
## Avoid
- dense starfield
- glitter noise
- messy sparkles
- too many small bright dots
- clutter around the face
- over-saturated rainbow glow
- cheap AI magic particles
## Light hierarchy
1. Face key light: keep identity readable.
2. Eye catchlight: make eyes beautiful and expressive.
3. Rim light: separate character from background.
4. Narrative light source: jar, cup, lantern, portal, moon, liquid, stage.
5. Sparse particles: only for spatial depth.
## Prompt snippets
```text
LIGHTFLOW: flowing iridescent light ribbons, amber liquid glow, wine-gold caustic reflections, prismatic rim light, glass-like reflections, soft jade/cyan highlights, sparse luminous particles only.
```
```text
Do not rely on many stars. No dense starfield, no glitter noise, no messy sparkles. Keep the face area clean and premium.
```
```text
The luminous reflections should feel like liquid wine, silk, glass, and prism light, not random glitter.
```
```
### references/storyboard-compositions.md
```md
# Storyboard Composition Reference
## Core principle
Multi-shot posters should feel like one cinematic scene with multiple camera cuts, not a random collage.
## Composition types
### 1. Split-panel close-up
- Left or upper panel: eye / half-face close-up.
- Right or lower panel: action gesture.
- Background: unified scene color and light.
Use when the user wants face/eye focus plus action.
### 2. Triptych key visual
- Top: eye close-up or symbolic object.
- Middle: main character action.
- Bottom: wide world / brand environment.
Use when the poster needs narrative richness.
### 3. Floating cinematic frames
- 2-3 translucent glass-like frames float inside one environment.
- Main face remains dominant.
- Frames share the same light direction and palette.
Use when the user wants anime opening / MV / trailer keyframe feeling.
### 4. Single-scene multi-exposure
- Same character appears in a close-up and a wider action silhouette inside one coherent scene.
- Avoid duplicated-face weirdness by describing panels/frames rather than random duplicates.
Use when the scene should feel poetic and less grid-like.
## Prompt snippets
```text
Use 2-3 integrated panels or floating cinematic frames, not a messy collage. Keep consistent lighting, color palette, and identity across all panels.
```
```text
Main panel: sharp face close-up with expressive eyes. Secondary panel: hand touching the glowing jar. Background panel: wide moonlit cellar world with soft depth of field.
```
```text
The panels should feel like sequential shots from one premium anime-inspired movie trailer, unified by the same amber and jade light.
```
## Negative constraints
```text
no messy collage, no random duplicated faces, no unrelated scenes, no inconsistent lighting, no comic strip clutter, no cheap split-screen effect
```
```
### scripts/generate_image2_series.py
```py
#!/usr/bin/env python3
"""
GPT-Image-2 image-to-image series generator template.
Copy this script into a project deliverables folder before use.
Fill API_KEY, REFERENCE_IMAGE, OUTPUT_DIR, and CARDS.
Do not store secrets inside the skill package.
"""
from pathlib import Path
import base64
import json
import time
import requests
API_URL = "https://ai.t8star.cn/v1/images/edits"
API_KEY = "" # Fill in project copy only, never commit secrets.
MODEL = "gpt-image-2-all"
SIZE = "1024x1792"
REFERENCE_IMAGE = Path("/absolute/path/to/reference.png")
OUTPUT_DIR = Path("./cinematic_keyframe_outputs")
COMMON_IDENTITY = """
Use the reference image only as a strict identity anchor, not as a composition or background reference.
Preserve the same face, eyes, skin tone, hairstyle, age impression, body proportion, and recognizable personal identity.
Do not invent a different person. Do not copy the original background, lighting, pose, or low-quality photo texture.
Identity priority is higher than costume, background, and visual effects.
""".strip()
COMMON_LIGHTFLOW = """
LIGHTFLOW: flowing iridescent light ribbons, amber liquid glow, wine-gold caustic reflections, prismatic rim light, glass-like reflections, soft jade/cyan highlights, subtle volumetric light, sparse luminous particles only.
Avoid relying on stars. No dense starfield, no glitter noise, no messy sparkles. Keep the face area clean and premium.
""".strip()
COMMON_NEGATIVE = """
NEGATIVE: changed identity, anime face, doll face, stiff portrait, static ID photo, static passport portrait, cluttered collage, dense starfield, glitter noise, too many sparkles, extra people, duplicated face, deformed hands, wrong Chinese characters, random letters, watermark.
""".strip()
CARDS = [
{
"id": "01_eye_closeup",
"copy": "眼中有光·杯里有山河",
"visual": "Create a premium vertical 9:16 cinematic close-up key visual poster. Focus on beautiful expressive eyes and face close-up. Use a realistic cinematic face, but borrow anime-inspired camera language. Main panel: extreme close-up of eyes with amber reflections. Secondary panel: hand touching a glowing artifact. Background panel: wide brand world with soft depth of field.",
},
]
def build_prompt(card):
return f"""
{COMMON_IDENTITY}
{card['visual']}
COMPOSITION: Use 2-3 integrated panels or floating cinematic frames, not a messy collage. Keep consistent lighting, color palette, and identity across all panels.
STYLE: realistic cinematic photography, luxury Chinese fantasy movie key visual, anime-inspired storyboard composition, shallow depth of field, rich framing, clean premium poster design.
{COMMON_LIGHTFLOW}
TEXT TO RENDER:
Add only this exact Chinese copy: "{card['copy']}".
The text must be clear, readable, correctly written, elegant, and integrated into the poster.
Do not add any other text.
{COMMON_NEGATIVE}
""".strip()
def generate_one(card):
if not API_KEY:
raise RuntimeError("Fill API_KEY in the project copy of this script before running.")
OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
prompt = build_prompt(card)
prompt_path = OUTPUT_DIR / f"{card['id']}_prompt.md"
prompt_path.write_text(prompt, encoding="utf-8")
image_data = REFERENCE_IMAGE.read_bytes()
files = {
"image": (REFERENCE_IMAGE.name, image_data, "image/png"),
"prompt": (None, prompt),
"model": (None, MODEL),
"size": (None, SIZE),
"n": (None, "1"),
}
headers = {"Authorization": f"Bearer {API_KEY}"}
resp = requests.post(API_URL, headers=headers, files=files, timeout=600)
if resp.status_code != 200:
raise RuntimeError(f"API failed {resp.status_code}: {resp.text[:1000]}")
data = resp.json()
b64 = data["data"][0].get("b64_json")
if not b64:
raise RuntimeError(f"No b64_json in response: {json.dumps(data)[:1000]}")
out_path = OUTPUT_DIR / f"{card['id']}.png"
out_path.write_bytes(base64.b64decode(b64))
meta_path = OUTPUT_DIR / f"{card['id']}_meta.txt"
meta_path.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
return out_path
def main():
manifest = []
for index, card in enumerate(CARDS, 1):
print(f"[{index}/{len(CARDS)}] generating {card['id']}...")
out = generate_one(card)
manifest.append({"id": card["id"], "copy": card["copy"], "path": str(out)})
time.sleep(1)
(OUTPUT_DIR / "series_manifest.json").write_text(json.dumps(manifest, ensure_ascii=False, indent=2), encoding="utf-8")
print(f"Done: {OUTPUT_DIR}")
if __name__ == "__main__":
main()
```
---
发布说明：这是可复用的 WorkBuddy Skill。完整 zip 包已在本地生成，可导入到 WorkBuddy skills 目录。