ip-diagram-creator

Personal IP Illustrated Creator

Use your IP persona to generate exclusive content illustrations.

Underlying principle: define the character first, then break down the content, and finally generate the illustration.

This Skill is not an avatar generator, nor a free-style image generator. Its goal is to turn user-authorized personal materials into a set of reusable minimalist hand-drawn illustrated characters, and then use this character together with an execution Agent to participate in content expression.

Recommended Runtime Environment

Recommended for use in an Agent environment that supports image reading, higher reasoning intensity, and image generation tools. This allows direct completion of image reading, character profiling, generation of three character assets, content illustration generation, and revisions.

If the current environment supports image_gen, GPT Image 2, or other image generation tools, images can be generated directly when the user explicitly says "generate / produce image / make the image directly." It is recommended to use higher reasoning intensity for character profiling, long-form article breakdowns, knowledge card design, and character consistency revisions.

If the current environment does not have image generation tools, do not pretend that images have already been generated. Instead, output a complete prompt that can be copied into an image generation tool, along with a revision prompt.

Environment Assessment

Before each start, first assess the current capabilities:

1. If there is an image generation tool:
   • When the user explicitly says "generate / produce image / make the image directly," you can generate directly.
   • Generate each image separately; do not combine multiple images into one, unless the user explicitly wants a card set overview.
   • Before calling, still construct a complete prompt to facilitate revisions and reuse.
2. If there is no image generation tool:
   • Do not directly claim that an image has been generated.
   • Output a complete prompt that can be copied into common image generation models.
   • Also output a revision prompt.
   • If the user needs multiple assets, output them separately 01_角色主锚图、02_角色规范说明图、03_动作_表情_小比例场景扩展图 three prompts.
3. If the current model or environment cannot read images:
   • First ask the user to supplement the appearance description, or request that a readable image be uploaded.
   • Do not fabricate a strongly similar character out of thin air.
4. If the user uses low reasoning or a non-image model:
   • Prioritize delivering a structured character extraction card, content confirmation card, and complete prompt.
   • Do not forcefully promise complex graphic-text layout, character consistency, or that images have been generated.

User-Facing Language

Use these terms with users:

• Minimalist hand-drawn illustrated character
• White-background hand-drawn illustration
• Action metaphor
• Amount of information in the image
• Long-form article illustration
• Graphic-account knowledge card
• Mobile poster
• Method breakdown diagram
• Storyboard story image
• PPT presentation mode
• Full-page PPT page image
• Director planning card
• page card
• Agents
• Execution Agent
• AI workforce
• Executor
• User
• Feedback source

Do not write internal aliases in user-visible titles, tags, descriptions, and recommendation texts. When expressing auxiliary roles, write them semantically as Agents, execution Agent, executor, user, failure case, risk warning, or feedback source.

In knowledge card mode, auxiliary execution Agents are not optional decorations. For any content involving steps, processes, methodologies, comparisons, cases, risks, or action tickets, assign 2-6 execution Agents by default, each respectively responsible for actions such as execution, feedback, error demonstration, card handling, risk marking, and result submission; the user IP is responsible for the main presentation, breakdown, annotation, or scheduling. Only for pure opinion posters, minimalist covers, or when the user explicitly requests "no little figures / no Agents" should they be reduced or removed.

Entry judgment

First determine which entry the user is at:

A. The user does not yet have a character

When the user provides their own photo, profile page screenshot, avatar, bio, account information, or says "help me make a personal IP illustrated character", first enter character profiling.

Read:

• references/identity-and-character.md
• references/visual-language.md

Do not ask the user to fill in a form. First extract information from existing materials; only ask one question when key information that blocks generation is missing.

B. The user already has a character

When the user already has a character main anchor image, character specification sheet image, action / expression / small-scale scene expansion images, character profile, avatar, or main visual, skip character profiling and go directly to content understanding.

Read:

• references/content-workflow.md
• references/modes-and-sizes.md

C. The user only makes assets

When the user only wants an avatar, action library, expressions, prop components, or visual components, handle it as a character asset task and do not enter content orchestration.

D. The user provides content but has no character

First remind them that character profiling is needed. If the user only wants to use a generic identity-less character, do not force the personal IP process of this Skill.

E. PPT presentation mode

When the user already has a character, or explicitly uses themselves / a certain IP character as the main visual, and requests PPT, courseware, livestream sharing pages, themed presentations, a full set of presentation pages, or page-by-page image generation based on an outline/verbatim script, enter PPT presentation mode.

Read:

• references/ppt-presentation-mode.md

If the IP character has not been confirmed, first return to character profiling. PPT presentation mode does not serve ordinary business reports without a personal IP premise.

PPT presentation mode must first output the full set of director planning cards; after the user confirms, make 1-2 sample pages; only after the samples are confirmed should generation proceed in batches. Do not skip director planning and directly generate images in batches.

Overall workflow

阶段 1：角色建档
用户照片 / 主页截图 / 简介 / 一段话
-> 自动提取身份和视觉锚点
-> 输出角色信息提取卡
-> 生成三张角色固定资产
   01_角色主锚图
   02_角色规范说明图
   03_动作_表情_小比例场景扩展图
-> 用户确认 / 返修
-> 固化角色档案

阶段 2：内容理解
长文 / 链接 / 截图 / 课程大纲 / 视频脚本 / 一句话主题
-> 判断输入源类型
-> 拆核心观点、认知转折、适合图解的锚点

阶段 3：视觉模式判断与出图方案推荐
判断手绘插图模式 / 知识卡模式 / PPT 演讲模式
-> 普通内容图：推荐图类型、尺寸、图内信息量、结构类型、角色动作、图内文字
-> PPT 演讲模式：先输出整套导演规划卡

阶段 4：确认卡 / 导演规划确认
普通内容图：输出内容确认卡
PPT 演讲模式：输出导演规划卡，用户确认页数、节奏、密度、人物出现频率

阶段 5：生成与返修
当前环境有图像生成工具时可直接生成
没有图像生成工具时输出可复制 prompt
-> QA 检查
-> 返修

Character profiling rules

Read:

• references/identity-and-character.md
• references/visual-language.md
• references/safety-and-assets.md

Character profiling only solves "who the user is, what they look like in subsequent illustrations, and how they move". Do not ask the user to provide knowledge points in advance.

Output to the user:

1. Character information extraction card: form of address, field, bio, temperament, recognition anchors, occupational props, prohibited zones.
2. Three character fixed asset prompts or images：
   • 01 Character main anchor image: the default main reference for subsequent ordinary content images. The character should be large, with hairstyle, glasses, clothing, posture, and temperament clear and stable.
   • 02 Character specification sheet image: rules, prohibited zones, color scheme, props, and basis for correction. Used for user confirmation, QA, and rework; not passed to the image generation model by default each time.
   • 03 Action / expression / small-scale scene expansion image: Reference for complex actions, expressions, and small-scale scene characters. It must include small-scale scene characters and a "protagonist + Agent collaboration" section, ensuring that recognition anchors are retained even when characters occupy only 10%-20% of the frame, and that scheduling, annotation, acceptance, and correction can also be handled in knowledge cards.
3. Confirm the issue: Ask the user to confirm whether the three assets can serve as long-term illustrated character assets.

Before character confirmation is completed, do not generate content illustrations.

After the character assets are confirmed, the reference priority is:

角色主锚图 > 动作/表情/小比例场景扩展图 > 角色规范说明图 > 原始照片 > 临时生成图 > 模型自由发挥

Subsequent image upload rules:

• Ordinary content images: by default, upload the character main anchor image.
• Hand-drawn illustration mode with characters becoming smaller in a large scene: upload the character main anchor image + action/expression/small-scale scene expansion image + the corresponding hand-drawn illustration reference image.
• Complex actions, multiple poses, multiple expressions: upload the character main anchor image + action/expression/small-scale scene expansion image.
• Character deviation, rework, or review: check the character specification explanation image and upload additional images if necessary, but do not treat it as the default main reference.
• Knowledge card mode: usually upload the character main anchor image; if the user IP needs actions such as explaining, pointing, annotating, or carrying cards, then add the action expansion image.

Content understanding rules

Read:

• references/content-workflow.md
• Long text or multi-paragraph content: focus on using the shot list rules within it
• One-sentence topic: use the one-sentence expansion rules within it

After content is input, first determine the input source; do not first ask the user to choose the scene, size, or amount of in-image information. Understand the content first, then recommend an image plan.

For long text, by default provide a shot list first, and do not directly generate all images. If the article is very short, 1-3 images may be recommended; ordinary long articles usually use 4-8 images, and should not easily exceed 9 images.

Do not directly throw a one-sentence topic to the image generation tool. First expand it into a confirmable illustration script, then generate a content confirmation card.

Image plan recommendations

Read references/modes-and-sizes.md。

If it is PPT presentation mode, also read references/ppt-presentation-mode.md。

First determine the visual mode:

• Hand-drawn illustration mode: Hand-drawn illustrated illustration on a white background, with a lot of white space, one main metaphor or one action logic. Characters are often small-scale action characters, and stability must be maintained through the character main anchor image and the small-scale scene expansion image.
• Knowledge card mode: A complete content container that can carry titles, modules, steps, risks, actions, key quotes, scores, screenshot annotations, course structure, etc. The knowledge card form may refer to samples, but should not be limited by samples; create a suitable form based on the content structure, publishing ratio, and reading scenario. The user IP is usually the presenter, decomposer, annotator, or scheduler; the execution Agent is by default interspersed in methods, steps, processes, comparisons, cases, risks, and action work orders, taking on execution, feedback, trial and error, carrying cards, submitting results, and marking risks.
• PPT presentation mode: A continuous page container. Used for livestream sharing, courseware, keynote speeches, and case reviews with a personal IP character premise. First create the full-set director planning card, then make a sample, then generate the full-page PPT page images in batches.

Each recommendation must include:

视觉模式：[手绘插图模式 / 知识卡模式]
首选：[图类型]
默认尺寸：[比例]
图内信息量：[画面更空 / 讲清楚重点 / 内容更完整 / 完整海报]
结构/形态：[路线图卡 / 白板讲解卡 / 行动工单 / 或为本内容定制的新形态]
角色分工：[用户 IP 如何参与核心动作；Agents / 执行者 / 用户等如何辅助]
辅助 Agent 分工：[哪些内容需要执行 Agent；每个 Agent 做什么动作；是否需要失败案例 / 用户反馈 / 风险提示]
出图效果：[用户能理解的结果描述]
推荐原因：[基于内容结构的判断]

还可以选：
1. [备选图类型]：[效果说明]
2. [备选图类型]：[效果说明]

Do not lock the recommendation. The user can change the image type, size, and amount of in-image information.

If it is PPT presentation mode, the recommended output changes to:

视觉模式：PPT 演讲模式
整套目标：[直播分享 / 课程课件 / 主题演讲 / 案例复盘]
默认尺寸：16:9
初步页数：[按内容判断]
风格方向：个人 IP 手绘演讲 PPT 风
导演规划：先输出每页页面类型、视觉权重、主视觉、图文比例、人物使用和 QA 风险
样张建议：先选 1-2 页代表页确认字体、人物和信息密度
推荐原因：[基于大纲和听众体验的判断]

Content confirmation card

Before generating content-type images, a content confirmation card must first be provided:

图类型：
尺寸：
图内信息量：
核心观点：
图内标题：
图内文字：
主画面隐喻：
角色动作：
辅助 Agent 分工：
建议元素：
必须保留的词：
可自由发挥的部分：
本图参考的角色资产：

Skill's responsibility is to first organize the text and structure within the image clearly, and after the image is produced, check the text accuracy, readability, and information density.

Image generation and prompt routing

If the current environment has an image generation tool:

• When the user explicitly asks to "generate / produce an image / directly make an image," you may generate directly.
• Before calling, you must still construct a complete prompt to facilitate rework and reuse.
• Generate each image separately; do not combine multiple images into one unless the user explicitly wants a card deck overview.

If the current environment does not have an image generation tool:

• Output a prompt that can be copied to common image generation models.
• At the same time, output a rework prompt.

Prompt writing

Read references/prompt-templates.md。

Each prompt must include:

• The confirmed personal IP character anchor.
• The character assets to be referenced this time: character main anchor image / action extension image / specification explanation image.
• White background, minimalist hand-drawn, negative space, restrained red-orange-blue annotations.
• The current image type and size.
• The core viewpoint, in-image text, main visual metaphor, character action, and auxiliary Agent division of labor in the content confirmation card.
• Hand-drawn illustration mode / knowledge card mode: prohibit too PPT-like, commercial poster, childish cute, complex background, and copying old images.
• PPT presentation mode: must look like a PPT page, but prohibit cheap templates, training handouts, infographic stacking, chaotic fonts, character drift, and a commercial poster feel.

QA and rework

Read references/qa-repair.md。

If it is PPT presentation mode, also read references/ppt-presentation-mode.md single-page QA, full-deck QA, and rework priority.

Check order:

1. Whether the character resembles the character main anchor image, and whether the recognition anchor is preserved; if it does not resemble the main anchor image, it is directly unqualified.
2. Whether the character participates in the core action, rather than standing beside it as decoration.
3. Whether the content accurately expresses the viewpoint confirmed by the user.
4. Whether the in-image text is accurate, readable, and not excessive.
5. Whether the style is still white-background hand-drawn, negative space, and restrained.
6. If it is a methodology / steps / process / comparison / case / risk / action work order type knowledge card, whether there is an execution Agent or action metaphor; if there is only the IP protagonist + text, it is judged as insufficient illustrative feel.
7. Whether hand-drawn illustration mode / knowledge card mode is too PPT-like, too commercial, too childish, or too refined; whether PPT presentation mode does not look like a PPT page, or looks like a cheap template, training handout, or infographic stacking.
8. Whether it copies the composition of an old image.
9. Whether user-visible text mistakenly writes internal code names; if so, change them to Agents, execution Agent, user, executor, feedback source, etc.

During rework, convert the user's language into specific modifications; do not reinvent the entire set of characters unless the user requests a redo.

Security and authorization boundaries

Read references/safety-and-assets.md。

By default, only process photos, homepage screenshots, profiles, account materials, and content materials explicitly authorized and provided by the user. Do not encourage users to upload portraits of others, private screenshots, or unauthorized reference images.

Do not put a character asset confirmed by a certain user into the public Skill package as a default asset. The public package may only contain general reference images, layout examples, and text explanations.