persona-pack

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Persona Pack

Persona包

Follow shared release-shell rules in:

```
postplus-shared
```
release-shell rules

Use this skill when a production workflow needs:

a creator persona recommendation
persona comparison options
a locked visual identity for repeated content
image prompt packs for batch generation

This skill is not for unconstrained character design.

请遵循以下共享发布框架规则：

```
postplus-shared
```
发布框架规则

当生产工作流需要以下内容时，可使用本技能：

创作者Persona推荐
Persona对比选项
用于重复内容的锁定视觉标识
用于批量生成的图像提示词包

本技能不适用于无约束的角色设计。

Fact Rule

事实规则

Personas must be built from observed benchmark evidence.

Use source-backed fields such as:

```
videoCreatorType
```
```
videoProtagonist
```
```
videoVisualStyle
```
```
videoHook
```
lane distribution
strong benchmark subsets
final report conclusions

Do not fabricate:

demographic precision not present in sources
audience fit claims without behavioral evidence
visual traits chosen only because they "look good"

If a detail is inferred rather than directly observed, label it as inference.

Persona必须基于已观测到的基准证据构建。

使用有来源支撑的字段，例如：

```
videoCreatorType
```
```
videoProtagonist
```
```
videoVisualStyle
```
```
videoHook
```
赛道分布
优质基准子集
最终报告结论

禁止编造以下内容：

来源中未提及的精准人口统计信息
无行为证据支撑的受众适配声明
仅因“看起来不错”而选择的视觉特征

若某一细节是推断得出而非直接观测到的，需标注为推断内容。

Default Workflow

默认工作流程

1. Start from strong benchmark subsets

1. 从优质基准子集入手

Do not define a persona from the full pool alone if strong benchmark subsets exist.

Prefer:

strong benchmark talking-head videos
lane-specific strong samples
pattern table summaries
final report conclusions

若存在优质基准子集，则不能仅从全部样本池中定义Persona。

优先选择：

优质基准真人出镜视频
特定赛道的优质样本
模式表格汇总
最终报告结论

Source Selection Rule

来源选择规则

Start from the active project's strongest benchmark artifacts.

If the task clearly belongs to one project or client folder, read from that folder first.

Do not assume one client folder is the default source base for all persona work.

Use the master table and shortlist to extract:

recurring creator types
protagonist descriptors
visual-style descriptors
strong benchmark handles and URLs

Do not let image-model aesthetics override benchmark-supported appearance patterns.

从当前项目的优质基准成果入手。

若任务明确属于某一项目或客户文件夹，优先读取该文件夹中的内容。

请勿默认将某一客户文件夹作为所有Persona工作的默认来源库。

使用主表格和候选列表提取：

重复出现的创作者类型
主角描述符
视觉风格描述符
优质基准账号及URL

请勿让图像模型的审美偏好覆盖基准支持的外观模式。

2. Extract persona evidence

2. 提取Persona证据

Before proposing any persona, collect:

common creator roles
common age feel or life-stage feel if directly implied
recurring wardrobe patterns
recurring environment patterns
recurring camera patterns
recurring speaking style
recurring authority style

Write these as evidence notes first.

在提出任何Persona之前，需收集以下内容：

常见创作者角色
若有直接暗示，常见的年龄感或人生阶段感
重复出现的着装模式
重复出现的环境模式
重复出现的镜头模式
重复出现的说话风格
重复出现的权威风格

先将这些内容整理为证据笔记。

3. Recommend the narrowest viable persona

3. 推荐最精准可行的Persona

The first persona should maximize:

fit with strongest lanes
repeatability across 10+ videos
visual consistency
credibility for the product

Avoid personas that are:

too broad
too aspirational
too polished and ad-like
unsupported by evidence

首个Persona应最大化满足以下要求：

与最优赛道的适配性
在10+条视频中的可复用性
视觉一致性
对产品的可信度

避免选择以下类型的Persona：

过于宽泛
过于理想化
过于精致、类似广告风格
无证据支撑

4. Produce a persona lock

4. 生成Persona锁定方案

Every chosen persona should be converted into a stable pack with:

```
personaId
```
```
archetype
```
```
sourceBasis
```
```
appearanceSystem
```
```
outfitSystem
```
```
environmentSystem
```
```
cameraSystem
```
```
voiceStyle
```
```
behavioralConstraints
```
```
negativeConstraints
```

This becomes the upstream source for image generation.

每个选定的Persona都应转换为一个稳定的包，包含：

```
personaId
```
```
archetype
```
```
sourceBasis
```
```
appearanceSystem
```
```
outfitSystem
```
```
environmentSystem
```
```
cameraSystem
```
```
voiceStyle
```
```
behavioralConstraints
```
```
negativeConstraints
```

这将成为图像生成的上游来源。

5. Separate direct observation from generation guidance

5. 区分直接观测内容与生成指引

Use two blocks:

```
Observed from benchmarks
```
```
Prompt translation
```

That prevents prompt language from being mistaken as factual research.

使用两个模块：

```
Observed from benchmarks
```
（基准观测内容）
```
Prompt translation
```
（提示词转换）

这样可避免提示词语言被误认为是事实研究内容。

Output Shapes

输出形式

Common outputs:

persona recommendation memo
3-way persona comparison
final chosen persona lock
image prompt pack
negative prompt pack

常见输出：

Persona推荐备忘录
三选一Persona对比
最终选定的Persona锁定方案
图像提示词包
反向提示词包

Persona Recommendation Memo

Persona推荐备忘录

Include:

recommended first persona
rejected alternatives
evidence basis
why this persona best fits the first 10 videos

需包含：

推荐的首个Persona
被否决的备选方案
证据依据
为何该Persona最适合前10条视频

Persona Lock

Persona锁定方案

Should include:

identity summary
repeated visual rules
what must stay fixed
what can vary
what to avoid

应包含：

身份摘要
重复视觉规则
必须固定的内容
可变动的内容
需避免的内容

Negative Constraints

反向约束

Always define what not to generate.

Common examples:

too much studio polish
luxury lifestyle look
founder keynote energy
generic influencer glam
overproduced ad lighting

Negative constraints are important because many image models drift toward polished ad aesthetics.

务必定义禁止生成的内容。

常见示例：

过度的工作室精致感
奢华生活风格外观
创始人 keynote 式气场
通用网红 glamour 风格
过度制作的广告灯光

反向约束十分重要，因为许多图像模型会倾向于生成精致的广告风格内容。

Source Basis Requirement

来源依据要求

Every persona recommendation must cite concrete support, such as:

strong talking-head lane distribution
creator-type counts
representative benchmark ids
visual-style descriptions
report conclusions

If the user asks for a persona that is not supported, state that clearly and provide:

closest evidence-backed version
what would need to be researched to support the requested persona

每一项Persona推荐都必须引用具体的支撑内容，例如：

优质真人出镜赛道分布
创作者类型数量
代表性基准ID
视觉风格描述
报告结论

若用户要求的Persona无证据支撑，需明确说明，并提供：

最接近的有证据支撑版本
若要支撑该请求的Persona，需要开展哪些研究