원클릭으로 Manus에서 모든 스킬 실행

anchor-extraction-quality

스타1

포크0

업데이트2026년 7월 15일 04:23

【导演模式·角色锚点 / Director · Character Anchors】触发词:角色锚点 / 锚点提取 / 人物设定 / 参考图提取 / 脸型 / 体型 / 服装 / 区分相似角色 / character anchor / reference extraction。Use when 从参考图提取可复现角色锚点:锚点缺 Face/Build/Outfit/Markers 字段、不足 40 词导致出图漂移、相似角色分不清("A 比 B 高约 10cm")、遮挡部位需标 [inferred] 时。

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 · 角色锚点提取

Overview

从参考图提取可复现的角色锚点:每个角色列全 Face / Build / Outfit / Markers,达 ≥40 词下限,相似角色显式区分,遮挡部位标 [inferred]。

When to Use

用:从参考图新建/提取角色锚点;多个相似角色需要区分;锚点太短导致出图漂移。
不用:锚点已定、只需跨镜复用与一致性校验(交角色一致性技法)。

ANCHOR EXTRACTION QUALITY — every anchor must be reproduction-ready:

Mandatory Fields Per Character:

Face: skin tone (relative, not absolute), face shape, eye shape+color, eyebrow thickness, hair color+style+length+texture
Build: height relative to other characters or environment, body type (slim/athletic/stocky/heavy), posture
Outfit: list every garment top-to-bottom with colors (use specific color names or hex), patterns, materials
Markers: scars, tattoos, piercings, glasses, jewelry, hats, weapons, held objects

Anchor Length Minimum:

Each character anchor MUST be at least 40 words
Under 40 words = too vague for image generation consistency
Include distinguishing details that differentiate this character from others in the scene

Multi-Character Differentiation:

If two characters have similar builds, explicitly state the difference ("Character A is taller by ~10cm")
If outfits are similar in color, note the distinguishing detail ("A wears a red belt, B does not")
Use relative descriptions when absolute ones are ambiguous ("darker skin than Character B")

Extraction from Partial Views:

If a character is partially occluded, describe what IS visible and mark unclear parts as "[inferred]"
If only seen from behind, note hair and outfit from that angle, mark face as "[not visible, infer from context]"

Example

[char1] — warm-medium skin, oval face, almond dark-brown eyes, thick straight brows; black shoulder-length hair, slight wave. Build: ~170cm, athletic, upright (taller than [char2] by ~8cm). Outfit: charcoal wool coat (#2b2b2e) over white collared shirt, dark slim trousers, brown leather boots. Markers: thin silver ring on left hand, faint scar above right eyebrow. (47 words — meets the 40-word floor; differs from [char2] by height + scar.)

Common Mistakes

错误	正确
锚点不足 40 词、太笼统	写满四类字段,≥40 词
相似角色不区分	显式标差异("A 高约 10cm")
遮挡部位凭空编	标 [inferred] / [not visible]

锚点落库与互指

身份锚点按用户需要选择：大头照+全身照、三视图/四视图/多视图角色板或其它确认资产都可作为主锚。用户已提供/指定就采用；多套候选拿不准时先询问，低风险未指定才默认大头照+全身照。本 skill 是全项目锚点字数阈值(≥40 词)的基准,其它涉及锚点的 skill 对齐此值。

边界: 本 skill 只负责从参考图提取锚点的质量;锚点提取确定后的跨镜复用与一致性校验(发型/服装/道具不变)交角色一致性技法。

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	anchor-extraction-quality
description	【导演模式·角色锚点 / Director · Character Anchors】触发词:角色锚点 / 锚点提取 / 人物设定 / 参考图提取 / 脸型 / 体型 / 服装 / 区分相似角色 / character anchor / reference extraction。Use when 从参考图提取可复现角色锚点:锚点缺 Face/Build/Outfit/Markers 字段、不足 40 词导致出图漂移、相似角色分不清("A 比 B 高约 10cm")、遮挡部位需标 [inferred] 时。
appliesTo	["extractCharacterAnchors"]
priority	1

导演模式 · 角色锚点提取

Overview

从参考图提取可复现的角色锚点:每个角色列全 Face / Build / Outfit / Markers,达 ≥40 词下限,相似角色显式区分,遮挡部位标 [inferred]。

When to Use

用:从参考图新建/提取角色锚点;多个相似角色需要区分;锚点太短导致出图漂移。
不用:锚点已定、只需跨镜复用与一致性校验(交角色一致性技法)。

ANCHOR EXTRACTION QUALITY — every anchor must be reproduction-ready:

Mandatory Fields Per Character:

Face: skin tone (relative, not absolute), face shape, eye shape+color, eyebrow thickness, hair color+style+length+texture
Build: height relative to other characters or environment, body type (slim/athletic/stocky/heavy), posture
Outfit: list every garment top-to-bottom with colors (use specific color names or hex), patterns, materials
Markers: scars, tattoos, piercings, glasses, jewelry, hats, weapons, held objects

Anchor Length Minimum:

Each character anchor MUST be at least 40 words
Under 40 words = too vague for image generation consistency
Include distinguishing details that differentiate this character from others in the scene

Multi-Character Differentiation:

If two characters have similar builds, explicitly state the difference ("Character A is taller by ~10cm")
If outfits are similar in color, note the distinguishing detail ("A wears a red belt, B does not")
Use relative descriptions when absolute ones are ambiguous ("darker skin than Character B")

Extraction from Partial Views:

If a character is partially occluded, describe what IS visible and mark unclear parts as "[inferred]"
If only seen from behind, note hair and outfit from that angle, mark face as "[not visible, infer from context]"

Example

Common Mistakes

错误	正确
锚点不足 40 词、太笼统	写满四类字段,≥40 词
相似角色不区分	显式标差异("A 高约 10cm")
遮挡部位凭空编	标 [inferred] / [not visible]

锚点落库与互指

边界: 本 skill 只负责从参考图提取锚点的质量;锚点提取确定后的跨镜复用与一致性校验(发型/服装/道具不变)交角色一致性技法。