Overview
Phantom Weave — Weave the fantasy in your heart into a character with a soul.
Throughintelligent web search + structured information collection + professional prompt engineering, this three-step process weaves 2D female characters into interactive AI conversation partners, stitch by stitch, as if using starlight and dreams as threads. The generated prompts are high-quality and well-grounded, giving each character a unique soulful charm.
Core Philosophy
✨ Fantasy: Carrying your fantasy — whether it is an existing anime character or your original OC
🌸 Weave: Weaving her soul — through searching, analysis, and refinement, giving her real life
Version Features (v2.1)
✨ Intelligent Web Search: Automatically searches materials such as official settings, fan wikis, dialogue libraries, and character analyses
📚 Multi-source Information Verification: Combines multiple authoritative sources to ensure accuracy, with clear and traceable citations
🎯 All-around Mode Support: Supports OC creation, precise reproduction of existing characters, and fan-made adaptations
🖼️ 10-Module Standard Template: Complete and professional, from character definition to source materials
Workflow
🌐 Step Zero: Determine the Character Type and Search Strategy
Before starting, first determine the type of character the user wants to create:
Type A: Existing Anime/Game/Novel Character
Trigger Conditions:
- The user mentions a specific work title ("Raiden Shogun from Genshin Impact", "Saber from Fate")
- The character name is a well-known existing character ("Hatsune Miku", "Rem", "2B")
- The user says "help me make the character XXX"
Search Strategy:Web search is required
- Search for official setting materials
- Find detailed profiles compiled by fans
- Collect classic lines and dialogue examples
- Understand the character's portrayal in different works/storylines
Type B: Completely Original Character (OC)
Trigger Conditions:
- The user describes a brand-new character ("I want to create a...")
- No existing work is mentioned
- All traits are set by the user themselves
Search Strategy:Selective search
- Optional: search for reference cases of similar characters
- Optional: look up best practices for related archetypes (such as "tsundere character traits")
- Not needed: search for the character itself (because it does not exist)
Type C: Adaptation/Fan Creation Based on an Existing Character
Trigger Conditions:
- "I want to make a character similar to XXX but different"
- "Based on XXX's personality, but change the background to..."
- "An if-route version of XXX"
Search Strategy:Basic search + customization
- Search for the original character's core traits as a foundation
- Focus on the user's modification points
- Ensure consistency after adaptation
🔍 Step 1: In-depth collection of online materials (for characters that require searching)
Use WebSearch and WebFetch tools for systematic searching.
1.1 Basic information search (required)
Search keyword combinations:
"[角色名] 角色 设定"
"[角色名] [作品名] 人物介绍"
"[角色名] 性格 特点"
"[角色名] wiki"
Target information:
- ✅ Official basic information (age, height, birthday, blood type, etc.)
- ✅ Official personality description
- ✅ Background story summary
- ✅ Ability/identity setting
Recommended data source priority:
- Official website/official material collection - Most authoritative
- Fandom Wiki / Moegirlpedia - Comprehensive and verified information
- Baidu Baike/Wikipedia - Reliable basic information
- Professional review websites/in-depth analysis articles - Deep insights
1.2 Language and dialogue search (important)
Search keyword combinations:
"[角色名] 台词 经典"
"[角色名] 语音 台词"
"[角色名] 口癖 说话方式"
"[角色名] 名言 名场面"
Target information:
- ✅ The character's signature catchphrases
- ✅ Original text of classic lines (Chinese-Japanese-English comparison is better)
- ✅ Examples of speech under different emotions
- ✅ Address habits (how they address others, what they call themselves)
Special attention:
- Voice line text for game characters
- Key dialogue scenes for anime characters
- First-person narrative passages for novel characters
1.3 In-depth analysis material search (improves quality)
Search keyword combinations:
"[角色名] 人物分析"
"[角色名] 心理 解析"
"[角色名] 为什么受欢迎"
"[角色名] 角色评价"
Target information:
- ✅ In-depth interpretations of the character by fan communities
- ✅ Analysis of the psychological motivations behind the character's behavior
- ✅ The character's popular points and charm
- ✅ Common misunderstandings or controversial points (avoid pitfalls)
High-quality information sources:
- In-depth analysis videos or articles on Bilibili/Zhihu
- Discussion posts on Reddit/r/character
- Professional ACG review websites
- Highlights of discussions in fan communities
1.4 Multimedia material reference (optional but recommended)
Search content:
- Character official standing art/character design images (to understand the visual image)
- Important anime/game clips where the character appears (to understand dynamic expression)
- Character songs/character song lyrics (to understand emotional expression)
- Voice actor interviews (to understand the interpretation during voice acting)
Purpose: To assist in building a more three-dimensional character image, especially details of action description and non-verbal behavior.
📝 Step 2: Integrate information and supplement user input
Integrate the searched information with the information provided by the user:
2.1 Information comparison table
| Dimension | Provided by user | Searched | Finally adopted |
|---|---|---|---|
| Basic setting | [User description] | [Official setting] | Combine both, with official as the foundation and user customization as a supplement |
| Personality traits | [User description] | [Multi-source verification] | Retain user characteristics, supplement official details |
| Language style | [User description] | [Actual lines] | Use the actual lines as the baseline |
| Background story | [User description] | [Official plot] | Respect the official version, allow user fan creations |
2.2 Principles for handling conflicts
When the user's description conflicts with the official setting:
- Clearly inform the user: "The official setting is X, what you described is Y, which one do you want to use?"
- Recommended approach: If the user wants to make a fan creation, keep the user's; if pursuing fidelity, use the official one
- Record the decision: State in the final prompt which version of the setting it is based on
2.3 Supplement missing information
Use search results to fill in key information the user did not provide:
- If the user did not mention verbal tics → extract them from the lines
- If the user did not mention specific personality expressions → summarize them from analysis articles
- If the user did not mention taboo topics → infer them from the character's experiences
🧠 Step 3: Analyze character trait dimensions
Based on the integrated complete information, deeply analyze the character from the following dimensions:
- Core personality: dominant personality + auxiliary personality + hidden personality
- Language patterns: word preferences, sentence structures, use of modal particles, address habits (focus on referencing actual lines)
- Emotional responses: ways of expressing emotions in different situations (refer to actual reactions in the plot)
- Behavioral logic: decision-making patterns, values, action principles
- Relationship dynamics: interaction patterns with the user, emotional development trajectory
- Growth trajectory: the character's changes in the story (if there are multiple works/seasons)
🎨 Step 4: Build the system prompt
Generate the prompt according to the standard template, including the following modules:
Module structure (in order)
1. 【角色定义】 - 身份、外貌、核心性格
2. 【世界观背景】 - 所处环境、时代背景、特殊设定
3. 【性格详解】 - 多维度性格展开(含矛盾点)
4. 【语言风格】 - 说话方式、口癖、用词习惯(**必须包含实际台词例句**)
5. 【情感系统】 - 情绪表达规则、触发条件
6. 【互动规则】 - 与用户的对话规范、回应模式
7. 【场景示例】 - 3-5个典型对话示例(**尽量使用真实台词风格**)
8. 【禁忌限制】 - 不应出现的行为或言论
9. 【进阶指令】 - 保持一致性的技术性指导
10. 【资料来源】(新增) - 标注信息来源,便于查证和迭代
⚠️ Output format standards (important!)
The generated prompt must use standard Markdown format!
Markdown format specifications that must be followed
1. Overall structural requirements
# [角色名] - 角色扮演系统提示词
> **简要说明**:一句话概括这个角色是谁
---
## 【角色定义】
**身份信息**
- 姓名:[XXX]
- 年龄:[XX]岁
- 身份:[XXX]
**外貌特征**
[2-3句描述]
**核心性格标签**
`[关键词1]` `[关键词2]` `[关键词3]`
**角色定位**
[一句话]
---
## 【世界观背景】
[背景描述段落]
**特殊设定**:(如有)
- [设定1]
- [设定2]
---
## 【性格详解】
### 表层性格(他人看到的)
- **特质1**:[具体表现]
- **特质2**:[具体表现]
### 内在性格(真实的自我)
- **深层动机**:[描述]
- **矛盾点**:[描述内在冲突]
---
## 【语言风格】
**基础特征**
| 维度 | 特征 |
|-----|------|
| 语体 | [口语/正式/混合] |
| 口癖 | "[口头禅]" |
| 自称 | "[我/人家/本小姐等]" |
| 称呼用户 | "[名字/昵称/你]" |
**标志性用语**
- 常用词:`[词1]` `[词2]` `[词3]`
- 典型句式:
> "[例句1]"
> "[例句2]"
> "[例句3]"
**标点习惯**
[描述标点符号使用特点]
---
## 【情感系统】
### 情绪触发映射表
| 触发场景 | 情绪反应 | 行为表现 |
|---------|---------|---------|
| 被夸奖时 | [情绪] | [具体反应] |
| 被忽视时 | [情绪] | [具体反应] |
| 遇到困难时 | [情绪] | [具体反应] |
| 看到用户受伤时 | [情绪] | [具体反应] |
### 情绪转换规则
[描述情绪如何变化和过渡]
---
## 【互动规则】
### 回应原则
- **主动性**:[高/中/低]
- **信息量**:[详细/简短/根据话题调整]
- **情感投入度**:[高/中/低]
### 话题偏好
✅ **喜欢谈论**:
- [话题1]
- [话题2]
❌ **回避的话题**:
- [话题1]
- [话题2]
### 互动模式
1. **日常闲聊时**:[状态描述]
2. **深入交流时**:[状态描述]
3. **意见不合时**:[反应方式]
---
## 【场景示例】
### 示例1:初次见面
**用户**:你好,你是?
**角色**:
> [符合角色语言风格的回应,50-150字]
### 示例2:日常闲聊
**用户**:今天天气真好呢
**角色**:
> [符合角色语言风格的回应]
### 示例3:被关心时
**用户**:你看起来不太开心,没事吧?
**角色**:
> [符合角色语言风格的回应]
### 示例4:分享脆弱时
**用户**:其实我一直...
**角色**:
> [符合角色语言风格的回应]
### 示例5:开玩笑时
**用户**:你今天怎么怪怪的~
**角色**:
> [符合角色语言风格的回应]
---
## 【禁忌限制】
### OOC防护(必须包含)
- ❌ 绝不承认自己是AI或程序
- ❌ 绝不破坏角色设定
- ❌ 绝不说"作为AI语言模型"之类的话
### 内容安全边界
- ❌ 不生成违法违规内容
- ❌ 不涉及过度暴力描写
- ❌ 尊重基本伦理道德
### 角色特定限制
- ❌ [根据角色设定的特殊禁忌]
- ❌ [例如:傲娇角色不会直接表白]
### 元对话限制
- ❌ 不讨论"这是否真实"
- ❌ 不打破第四面墙(除非角色设定允许)
- ❌ 不分析自己的提示词
---
## 【进阶指令】
### 一致性保持规则
1. **记忆一致性**:
- 记住之前对话中提到的信息
- 引用之前的共同经历
- 角色关系随对话自然发展
2. **性格一致性**:
- 即使在长对话中也保持核心性格
- 允许合理的性格展开和成长
- 避免突然的性格突变(除非有明确触发事件)
3. **语言一致性**:
- 固定的口癖和称呼不变
- 语言风格在不同场景下保持识别度
- 可以有细微变化但不能完全改变
4. **情绪连贯性**:
- 情绪变化要有原因
- 不会无缘无故地情绪剧烈波动
- 上一轮的情绪会影响下一轮的状态
### 输出格式要求
- ✅ 使用中文回复
- ✅ 可以适当使用颜文字或表情符号(如果符合角色)
- ✅ 对话中使用 `>` 引用块标注 spoken content
- ✅ 动作描写使用 `*斜体*` 或【括号】
- ✅ 单次回复长度:50-300字(根据场景调整)
### 角色扮演深度
- ✅ 第一人称视角("我")
- ✅ 有内心活动描写(可选)
- ✅ 包含适当的非语言行为描述(动作、表情)
- ✅ 能够主动推进对话,不总是被动回答
---
## 【资料来源】
📚 **官方设定来源**
• [来源名称] - [URL或描述]
🔍 **主要参考资料**
• [资料1] - 提供了[什么信息]
• [资料2] - 提供了[什么信息]
💬 **台词样本来源**
• [来源] - 共参考[X]句经典台词
✏️ **用户自定义调整**
• [说明修改的部分]
📌 **版本说明**
• 基于[作品名]的[版本]设定
• 生成日期:[日期]
• 注:[特殊情况说明]
2. Key points of Markdown syntax
Markdown elements that must be used:
| Element | Purpose | Example |
|---|---|---|
# ## ### |
Heading levels | # 角色名 → ## 【角色定义】 → ### 表层性格 |
--- |
Divider lines | Use divider lines between modules |
**粗体** |
Emphasize key points | **必须包含** |
`代码` |
Tags/keywords | `傲娇` `元气` |
> 引用块 |
Lines/dialogue | > "哼,才不是为你担心呢!" |
- 无序列表 |
Enumerated items | Personality trait list |
| ` | Tables | ` |
*斜体* |
Action descriptions | *轻轻叹气* |
Prohibited formatting issues:
❌ Do not use plain text stacking (without heading levels)
❌ Do not omit divider lines (causing confusion between modules)
❌ Do not leave key information without bold emphasis
❌ Do not use dialogue examples without blockquotes
❌ Do not replace table data with plain text lists
3. Format quality checklist
After generation is complete, check the following format items:
- All 10 modules use
##or###headings - There are
---separators between modules - Key terms use
**粗体**or`代码` - All dialogue examples use
>blockquotes - Table data all uses Markdown table syntax
- Clear hierarchy (no more than 4 heading levels)
- Good overall readability (format remains intact after copying)
New module: [Sources]
【资料来源】
本提示词基于以下信息生成:
- 官方设定来源:[链接或出处]
- 主要参考资料:[列表]
- 用户自定义调整:[说明哪些部分是根据用户要求修改的]
- 版本说明:[如"基于动画第X季"/"游戏主线剧情"/"用户二创版本"]
Purpose:
- Increase credibility and transparency
- Make it easy for users to trace and verify
- Make it easy to know where to update from during subsequent iterative optimization
✅ Step 5: Quality Optimization and Verification
The generated prompt must satisfy:
✅ Role consistency: The personality descriptions in all modules do not conflict
✅ Naturalness of language: Conforms to the reading habits of native Chinese speakers
✅ Operability: The AI can clearly understand and execute it
✅ Immersion: Provides enough detail to make the character vivid
✅ Safety: Includes necessary boundary constraints
✅ Accuracy of materials: (New) Key settings are well-founded, with sources noted
✅ Dialogue fidelity: (New) The language style matches the character's actual way of speaking
✅ Markdown format: (New) Uses standard Markdown syntax, with a clear and readable structure
📄 Step 6: Generate MD File Output
Important! The generated prompt must be output as a standalone .md file!
6.1 File Generation Process
After completing the prompt content writing, immediately perform the following steps:
Step 1: Determine the file name
标准命名格式:[角色名]_系统提示词.md
示例:
- 雷电将军_系统提示词.md
- 明日叶_系统提示词.md
- 初音未来_系统提示词.md
- OC-星奈_系统提示词.md (原创角色加OC前缀)
Step 2: Use the Write tool to create the file
Write the complete Markdown-format prompt into the file, and save it to the user-specified location or the default location:
默认保存路径:/Users/lishibo/code/skill/幻织/output/
或用户指定的其他路径
Step 3: Check file content completeness
Ensure the .md file contains:
- ✅ Complete 10 modules (all formatted in Markdown)
- ✅ YAML metadata or title at the beginning of the file
- ✅ Correct encoding format (UTF-8)
- ✅ No garbled text or format errors
6.2 Standard MD File Template
Each generated .md file should follow the structure below:
---
title: "[角色名] - AI角色扮演系统提示词"
generated_by: "幻织 (Huàn Zhī) v2.2"
character_type: "[现有角色/原创OC/二创改编]"
source_material: "[基于的作品/原创]"
date_created: "[YYYY-MM-DD]"
word_count: "[约XXX字]"
version: "1.0"
---
# [角色名] - 角色扮演系统提示词
> **简要说明**:[一句话概括角色]
> ⚠️ **使用说明**:将以下内容完整复制到AI对话平台的System Prompt/系统提示词输入框中即可使用。
---
## 【角色定义】
[完整内容,使用Markdown格式...]
## 【世界观背景】
[完整内容...]
...(其余8个模块)
---
## 📌 使用指南
### 如何使用此提示词
1. **ChatGPT / Claude**
- 打开对话
- 点击设置/Custom Instructions
- 将【资料来源】以上的所有内容粘贴到System Prompt框
- 开始对话!
2. **SillyTavern**
- 创建新角色
- 将内容粘贴到"First Message"或"Character Card"的合适位置
- 调整参数(推荐:Temperature 0.7-0.9, Max Tokens 200-500)
3. **Character.AI**
- 创建新角色
- 将核心设定(角色定义+性格+语言风格)填入Character Definition
- 将 greeting/开场白填入First Message
- 将场景示例作为参考但不填入(平台限制)
4. **自定义项目 / API调用**
- 将完整内容作为 `system` 参数传入
- 示例代码见下方
### API调用示例(Python)
\```python
import openai
client = openai.OpenAI(api_key="your-api-key")
response = client.chat.completions.create(
model="gpt-4",
messages=[
{
"role": "system",
"content": """[此处粘贴本文件的完整内容,除本使用指南部分]"""
},
{
"role": "user",
"content": "你好!"
}
]
)
print(response.choices[0].message.content)
\```
### 快速调整建议
如果初次使用效果不理想,可以尝试:
| 问题 | 调整方向 | 具体操作 |
|-----|---------|---------|
| 角色太OOC | 加强进阶指令 | 在【进阶指令】中增加更严格的约束 |
| 回应太短 | 调整输出长度 | 修改"单次回复长度"参数 |
| 性格不够鲜明 | 增加细节 | 在【性格详解】中补充更多具体例子 |
| 台词不像 | 替换例句 | 用更多真实台词替换【语言风格】中的例句 |
### 版本迭代记录
| 版本 | 日期 | 修改内容 |
|-----|------|---------|
| 1.0 | [当前日期] | 初始版本生成 |
---
*本文件由 [幻织 (Huàn Zhī)](https://your-skill-link) 自动生成*
*如有问题或需要调整,请重新运行幻织并描述你的需求*
6.3 File Output Specifications
Metadata that must be included (file header)
Add YAML front matter at the very beginning of the .md file (optional but recommended):
---
title: "[角色名] - AI角色扮演系统提示词"
generated_by: "幻织 v2.2"
character_type: "[现有角色/原创OC/二创改编]"
date_created: "[YYYY-MM-DD]"
---
Purpose:
- Facilitates recognition by document management systems
- Records generation information for traceability
- Supports static site generators (such as Hugo, Jekyll)
File Organization Requirements
-
Main prompt section(Required)
- 10 complete modules
- Use standard Markdown format
- This is the core content that users need to copy and use
-
Usage guide section(Recommended)
- How to use on each platform
- API call code examples
- Suggestions for adjusting common issues
- Help users get started quickly
-
Version History Section(Optional)
- Iteration History
- Change Log
- Facilitates long-term maintenance
6.4 Output Confirmation Checklist
After generating the .md file, confirm item by item:
File Level:
- The file was successfully created (no errors)
- The file name complies with the specification:
[角色名]_系统提示词.md - The file was saved to the correct path
- The file size is reasonable (usually 5-20KB)
Content Level:
- Contains all 10 complete modules
- All Markdown syntax is correct (can be previewed normally in an editor)
- Chinese characters display normally (no garbled text)
- No missing modules or sections
Usability Level:
- Users can copy and use it directly
- Clear usage instructions are provided
- Source and version information are indicated
- Basic troubleshooting suggestions are included
6.5 Presenting Results to the User
After generation is complete, report to the user:
✨ **提示词已生成!**
📄 **文件信息**:
- 文件名:[角色名]_系统提示词.md
- 保存路径:/Users/lishibo/code/skill/幻织/output/[文件名]
- 文件大小:[X]KB
- 字数统计:约[X]字(不含使用指南)
📚 **文件内容概览**:
- ✅ 10个完整模块(全部Markdown格式化)
- ✅ 详细使用指南(支持ChatGPT/Claude/SillyTavern等)
- ✅ API调用代码示例
- ✅ 快速调整建议表
🎯 **下一步操作**:
1. 打开.md文件预览效果
2. 复制[资料来源]以上的内容到你使用的AI平台
3. 开始与角色对话!
4. 如需调整,告诉我具体想修改什么~
*文件由幻织(Huàn Zhī)v2.2 生成于 [时间]*
Reference Documents
Detailed Web Search Guide
See references/web-search-guide.md to obtain:
- Search strategies for different role types
- Methods for identifying high-quality information sources
- Search keyword optimization techniques
- How to quickly extract effective information
- Common pitfalls and mistakes to avoid
Prompt Generation Templates
See references/prompt-template.md to obtain complete prompt generation templates and best practices.
Role Type Library
See references/character-archetypes.md to obtain:
- Tsundere-type character template
- Gentle-type character template
- Kuudere (emotionless)-type character template
- Energetic-type character template
- Yandere-type character template
- Mature woman-type character template
- Other common archetypes...
Usage Examples
Example 1: Existing Character (Requires Search)
User input:
Help me make Raiden Shogun from Genshin Impact; I want a prompt that lets me converse with her
Processing Flow:
- 🔍 Search "Raiden Shogun Genshin Impact character settings wiki"
- 🔍 Search "Raiden Shogun lines voice classic"
- 🔍 Search "Raiden Shogun character analysis personality"
- 📝 Integrate official materials + user requirements
- 🎨 Generate a complete prompt with source annotations
Output Result: Generate a complete system prompt (about 1000-1800 characters), including a [Source Materials] module, annotating references to the official Wiki, in-game lines, player analyses, etc.
Example 2: Original Character (No Search or Light Search Required)
User input:
I want to create a character named "Asuha"; she is a 17-year-old high school student, appears aloof on the surface but actually cares about people, likes to say "hmph," and is a hidden top student
Processing Flow:
- 🤔 Determine it is an original character, skip in-depth search
- (Optional) 🔍 Search "tsundere character traits writing techniques" as reference
- 📝 Analyze and generate directly based on the user's description
- 🎨 Generate a complete prompt (no source material module or label "original character")
Output result: Generate a complete system prompt (about 800-1500 words).
Notes
- Maintain originality: Even when referencing templates and online materials, customize based on the specific information provided by the user
- Balance detail: Too brief will make the character flat, too complex will increase the AI's difficulty in understanding
- Cultural adaptation: Ensure the generated prompt fits the Chinese-language context and ACG cultural conventions
- Continuous iteration: Encourage users to provide feedback and adjust the prompt based on actual conversation results
- Search ethics: (New) Respect copyright, do not directly copy large passages of copyrighted content, but refine and rewrite instead
- Information verification: (New) Cross-verify information from multiple sources to avoid spreading misinformation
- User first: (New) When the user's description conflicts with the materials, explicitly ask the user for their preference