원클릭으로 Manus에서 모든 스킬 실행

director-prompt-engineering

스타1

포크0

업데이트2026년 7월 10일 13:45

【导演模式·提示词结构 / Director · Prompt Structure】触发词:提示词结构 / 七字段 / 提示词模板 / 怎么写提示词 / 镜头+灯光+构图+风格 / prompt structure / 7-field / prompt template。Use when 提示词字段乱序、丢字段导致出图不稳,需要一条规范可复现的成品提示词时:按七字段顺序拼装(主体动作 → 角色引用 → 场景 → 镜头相机 → 光照 → 构图 → 风格情绪),120 词内,负向按需补充。

설치

Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.

Manus에서 실행

출처

2799662352

2799662352/ai-image-master

GitHub 저장소 열기 Creator 저장소 보기

다운로드

Manus에서 실행

导演模式 · 提示词结构(7 字段)

Overview

把每条提示词按固定 7 字段顺序拼装(主体动作 → 角色引用 → 场景 → 镜头相机 → 光照 → 构图 → 风格情绪),≤120 词,前置重点字段。

When to Use

用:需要一条规范、可复现的成品提示词;字段乱序/丢字段导致出图不稳。
不用:多镜连续性/叙事排序(→ continuity / narrative 系列);纯锚点提取(→ anchor-extraction)。

PROMPT STRUCTURE — follow this order for every panel prompt:

Subject + Action: "[char1] reaches for a door handle" (use character tags, not full descriptions)
Character Reference: use [char1] [char2] tags defined in the global section (never inline full appearance per panel)
Scene/Environment: location, weather, time of day
Shot Type + Camera: "medium shot, eye-level, 50mm lens"
Lighting: direction, quality, color temperature ("warm golden hour side-light from left")
Composition: "rule of thirds, subject at left intersection, depth of field blur on background"
Style/Mood: art style, color palette, emotional tone

Prompt Hygiene:

Write in English, present tense, descriptive noun phrases
Front-load the most important elements (image generators weight early tokens higher)
Never use subjective words alone ("beautiful", "amazing") — always pair with concrete descriptors
Maximum 120 words per panel prompt; beyond that generators lose coherence

Negative Prompt(默认正向,按需补充):

默认正向描述:画质要求写成正向(sharp focus, correct anatomy, five clear fingers, natural proportions),而非堆"不要…"清单。
仅当目标模型有独立负向字段且确需时补充:blurry, deformed, bad anatomy, extra limbs, watermark, text;人像可加 cross-eyed, asymmetric face,建筑可加 impossible geometry。

Example

Assembled in 7-field order (≤120 words): [char1] reaches for a brass door handle, leaning in — [char1] — dim Victorian hallway, dusk — medium shot, eye-level, 50mm — warm tungsten side-light from left, soft, 3200K — rule of thirds, subject at left intersection, shallow DoF on background — muted sepia palette, tense, noir mood. (按需)Negative: blurry, deformed, bad anatomy, extra limbs, watermark, text, cross-eyed.

Common Mistakes

错误	正确
字段乱序、重点放末尾	按 7 字段顺序,重要元素前置
每格内联重描角色外貌	用 [char1] 标签引用全局定义
默认堆负向词	默认正向,负向按需补充

이 저장소의 다른 Skills

같은 저장소

director-orchestrator

2799662352/ai-image-master

【导演模式·总调度 / Director · Orchestrator】复杂镜头设计的调度器:由 catimation-image / catimation-video / film-studio 在专业·制片级任务中加载。按 13 维电影摄制框架把镜头写成「物理可复现参数优先于情绪形容词」的结构化文本 (非 JSON)提示词,并按维度路由 director-* / storyboard-* 技法。Use when 任务需要复杂镜头设计、专业构图与光影协同、多维提示词优化、参考图分析复刻, 或多镜连续性设计;简单生成请求不触发本技能。

2026-07-161

film-studio

2799662352/ai-image-master

端到端「做电影/做片子」的制片总编排器,仅用于真正的多镜成片项目——把一个想法/故事一条龙串成成片:概念 → 剧本 → 分镜镜头表 → 角色与场景锚点 → 逐镜出图 → 图生视频 → 配音配乐 → 拼接 → 体检交付,每阶段带审批门(gate), 并要求每阶段实际调用本地工艺 skill、记录回执,而不是只点名或凭记忆重写。单镜等非成片请求不触发本技能 (那是视频入口快速/标准模式的事,规格同样不降)。触发词:做电影、做短片、拍一部、成片、一条龙、从剧本到成片、制片、宣传片、微电影、short film、full production、 film pipeline。

2026-07-161

sd2-pe

2799662352/ai-image-master

Seedance 2.0 video prompt engineering base (八大要素, path A/B routing, asset binding, 12-item coverage, convergence weights, progression arc, verified references). Auto-trigger on related prompt work: writing / rewriting / optimizing a Seedance or 视频提示词; 图生视频 / 文生视频 / 视频编辑 / 延长 prompt drafts; pasted drafts to improve; Seedance 全能参考 or binding-syntax questions; and the video entry prompt_engineered step before the generate tool. Skip for non-video copywriting or coding.

2026-07-161

seedance-cinematic-format

2799662352/ai-image-master

Seedance 12-item completeness and media-profile leaf for complex prompt work. Load with sd2-pe for path B, multishot, mixed-media, or expanded director / reference work; path A may skip it. Provides format-free content definitions, live-action/2D/3D profiles, language defaults, and checks. No routing or QA.

2026-07-161

seedance-video-craft

2799662352/ai-image-master

用 Seedance 2.0 满血版出片的深度实战 skill——由视频入口在**复杂 Seedance 任务** 中加载:多模态参考配比(9图/3视频/3音频)、多镜叙事、视频编辑/延长、复杂动作、口播/广告片、高质量商业交付、爆款体检。基于字节 Seedance 2.0 技术报告 (统一多模态音视频联合生成,4-15s,原生音频+口型,主体控制/运动操控/风格迁移/ 视频延长),给出全能参考配方、模型对齐提示词补充、爆款体检自评(改编自 Higgsfield Virality Predictor)与广告/短视频模式分类。简单单镜生成不触发本技能。

2026-07-161

storyboard-grid-to-seedance

2799662352/ai-image-master

用图像模型生成或使用 3×3/4×4/16 格故事板、镜头总览板、电影美术设定板，并在 Seedance 中把它们设为低约束氛围参考、让文字提示词主导成片时使用。触发场景:一张图出整套分镜、九宫格/多宫格、grid storyboard、富文本美术设定图、把故事板整图或拆格图喂给视频、提示词为主、降低分镜约束并保留导演发挥、舞蹈/动作序列网格、漫画页转视频。

2026-07-161

name

director-prompt-engineering

description

导演模式 · 提示词结构(7 字段)

Overview

把每条提示词按固定 7 字段顺序拼装(主体动作 → 角色引用 → 场景 → 镜头相机 → 光照 → 构图 → 风格情绪),≤120 词,前置重点字段。

When to Use

用:需要一条规范、可复现的成品提示词;字段乱序/丢字段导致出图不稳。
不用:多镜连续性/叙事排序(→ continuity / narrative 系列);纯锚点提取(→ anchor-extraction)。

PROMPT STRUCTURE — follow this order for every panel prompt:

Subject + Action: "[char1] reaches for a door handle" (use character tags, not full descriptions)
Character Reference: use [char1] [char2] tags defined in the global section (never inline full appearance per panel)
Scene/Environment: location, weather, time of day
Shot Type + Camera: "medium shot, eye-level, 50mm lens"
Lighting: direction, quality, color temperature ("warm golden hour side-light from left")
Composition: "rule of thirds, subject at left intersection, depth of field blur on background"
Style/Mood: art style, color palette, emotional tone

Prompt Hygiene:

Write in English, present tense, descriptive noun phrases
Front-load the most important elements (image generators weight early tokens higher)
Never use subjective words alone ("beautiful", "amazing") — always pair with concrete descriptors
Maximum 120 words per panel prompt; beyond that generators lose coherence

Negative Prompt(默认正向,按需补充):

默认正向描述:画质要求写成正向(sharp focus, correct anatomy, five clear fingers, natural proportions),而非堆"不要…"清单。
仅当目标模型有独立负向字段且确需时补充:blurry, deformed, bad anatomy, extra limbs, watermark, text;人像可加 cross-eyed, asymmetric face,建筑可加 impossible geometry。

Example

Common Mistakes

错误	正确
字段乱序、重点放末尾	按 7 字段顺序,重要元素前置
每格内联重描角色外貌	用 [char1] 标签引用全局定义
默认堆负向词	默认正向,负向按需补充