comic-ip-onboarder

Comic IP Onboarder

Onboard a new user IP into a comic infographic workflow. This skill is not responsible for directly replacing an existing IP skill, nor does it modify the character rules of an existing skill; it only produces an independent, confirmable, portable IP profile package.

First Hard Command

You must use imagegen's image generation capability to actually generate the final images. The final delivered images may only be image files generated by imagegen; it is absolutely forbidden to use HTML, CSS, SVG, canvas, webpage screenshots, slide screenshots, collage canvases, or any front-end rendering method to generate, fake, or substitute for the final images.
If the current execution environment or the confirmed tool in use is Codex, you must use Codex's default image generation capability to complete the final images; do not switch to HTML, CSS, SVG, canvas, webpage screenshots, slide screenshots, or any front-end rendering method as a substitute.
If this process creates, rewrites, or hands off any new downstream SKILL.md, the first article of the generated main document must also include this section's hard command verbatim; it must not be placed in an appendix, ordinary notes, or optional suggestions.

Core Boundaries

Do not modify any existing IP skill unless the user explicitly requests it.
Do not write new character details into other skills' SKILL.md。
Existing IPs may be used as structural references: reuse only the profile format, quality-check checklist, and confirmation process, not the appearance.
If the source is a real person's photo, make it a stylized IP inspired by the photo; do not pursue a realistic portrait.
Visual reference images must comply with the First Hard Command and use imagegen to actually generate images; do not treat HTML, CSS, SVG, canvas, webpage screenshots, or slide screenshots as final IP reference images.
Only reference images confirmed by the user may enter the final profile folder. Failed images and unconfirmed images must not be saved as reference assets.
All prompts delivered to the user for copying or for image generation must be written in Chinese, including front-view confirmation image prompts, three-view prompts, expression pack prompts, prompt capsules, and negative prompts. Only file paths, file names, and fixed technical identifiers may remain in English.
Do not leave the responsibility for moving forward to the user. Each time a stage is completed, you must proactively ask about the next step with a brief question and provide recommended options.
The first front-view solid-color confirmation image must be a complete full-body image. Even if the user provides only a headshot, half-body photo, or three-quarter photo, you must not output a half-body confirmation image; you need to conservatively complete the lower body, shoes, or bottom structure, and list the inferred parts under "Places you need to confirm."
When the source image is not a full-body photo, proactively suggest that the user add a full-body photo to improve accuracy; however, do not stop the process just because there is no full-body photo. If the user does not add an image for the time being, continue making conservative inferences based on the existing reference.
The expression and action pack must be generated based on the confirmed three-view sheet. If there is no three-view.png or the user has not confirmed the three-view sheet, do not generate any single expression or action image. Each of the 8 images, and every generation, must use this three-view sheet as an image reference input; it cannot be referenced only in the first image or the first round.
Every expression and action prompt must explicitly require: “Please use the uploaded three-view as the character reference for the same IP, strictly keep the head-to-body ratio, overall silhouette, facial features, eye rules, clothing, accessories, color anchors, and line style consistent; only change the expression, action, and props in this image.”
The number of expression and action sheets must be fixed at 8; do not generate 6, 9, 12, or any other number.
The expression and action sheets must be generated and output one by one, and each must be an independent image. It is forbidden to first generate a single collection, contact sheet, nine-grid, collage, large sticker sheet, or multi-action large canvas and then split it.
The 8 expression and action images must keep the same background color, the same character proportions, and the same set of lines and color scheme; each image only changes the expression, action, and optional props.

When to use

Use this skill when the user provides or describes a new IP and wants it to become a reusable character:

User photo or selfie -> stylized comic IP
Plush toy or figure -> flat hand-drawn character profile
Avatar or mascot -> IP DNA, three-view, expression and action sheets
Companion pet or object -> companion character DNA and appearance frequency rules
Existing successful IP skill -> extract the profile format for use with a new user IP

Do not use this skill to generate ordinary article illustrations. After the IP profile is confirmed, hand it off to the downstream comic infographic skill.

Quick invocation phrases

The user can directly copy any of the following sentences to start:

调用 comic-ip-onboarder，把这张图做成可复用漫画 IP。先输出 IP_DNA 草案和第一张正面全身纯色确认图提示词，所有提示词都用中文。注意：不管原图是半身还是大半身，第一张确认图都必须是完整全身。

使用 comic-ip-onboarder，基于我上传的照片创建一个漫画 IP 档案。请先分析身份锚点，再主动询问我是否补充全身照；如果我暂时不补，也请先保守生成完整全身的正面确认图提示词。

调用 comic-ip-onboarder，为这个新 IP 建档：先做 IP_DNA.md 草案，再做正面全身确认图、三视图和表情动作包。每完成一步都主动问我下一步。

调用 comic-ip-onboarder，基于已确认的三视图继续生成这个 IP 的表情动作参考包。必须生成 8 张独立单张图片，不要生成合集再切分；每张表情图都必须以上传的三视图作为同一角色参考，保持背景颜色和人物一致性。

Required reading

references/IP_DNA_TEMPLATE.md: Must be read before drafting any new IP DNA.

Output conventions

Each new IP should ultimately have a profile package created or planned like this:

<project-or-skill>/references/ip-profiles/<profile-id>/
├── IP_DNA.md
├── front-full-body.png
├── three-view.png
├── expressions/
│   ├── 01-*.png
│   ├── 02-*.png
│   ├── ...
│   └── 08-*.png
└── companion/
    ├── companion-DNA.md
    └── companion-three-view.png

IP_DNA.md is the first deliverable. Images are added only after the user confirms each stage.

Proactive advancement rules

This skill must advance the process like a project manager. Do not just say “done” at the end of a stage, and do not make the user guess the next step.

At the end of every staged output, it must include:

下一步我建议：<推荐动作>
请你确认：<一个清晰问题>
可选方向：
1. <推荐选项>
2. <备选选项>
3. <需要时才给第三个选项>

Questions should be short, asking only about what truly needs the user's judgment at the current stage. By default, put the recommended option first.

Proactive questioning rules after a stage ends:

After completing the source image analysis, ask the user whether they accept the identity anchors, and recommend “generate a front full-body confirmation image first.”
After completing the IP_DNA.md draft, ask which identity anchors need to be modified, and recommend “after confirmation, generate a front full-body confirmation image.”
If the source image is not a full-body photo, after completing the source image analysis, proactively remind: “For greater stability, it is recommended to add a full-body photo; if you do not have one for now, I will first conservatively complete it into a full-body character.” Then ask whether to add an image or continue.
After completing the front full-body confirmation image, ask whether this image passes, and recommend “after passing, generate the three-view.”
After completing the three-view, ask whether the three-view passes, and recommend “after passing, use this three-view as the image reference to generate 8 independent single expression and action reference images.” The user must be reminded: the expression and action sheets will use the uploaded three-view as the character reference and will not rely only on text descriptions; the expression sheets must be generated one by one, without splitting a collection.
After completing the expression and action sheets, ask whether a companion character is needed or whether to directly package and hand off, and recommend “first package the current IP profile.”
After completing the final archive package, ask the user for the next task: which article it is for, whether to install it into the downstream comic skill, or whether to create a new IP for another user.

If the next step depends on user aesthetic confirmation, such as whether the character looks right, whether the hairstyle is correct, or whether the clothing is retained, do not automatically skip confirmation and continue generating subsequent images. Ask proactively and provide suggestions for the next step.

Workflow

Read the source images, character descriptions, or existing reference assets provided by the user.
Assess the completeness of the source: full body, mostly full body, half body, headshot, or partial. If it is not full body, proactively suggest that the user supplement a full-body photo; if the user does not supplement it for the time being, continue conservatively completing the full-body character based on the existing reference, and list the inferred parts as items to be confirmed.
Read references/IP_DNA_TEMPLATE.md, draft IP_DNA.md。
The first-round draft must use these four parts for user confirmation:
- 保留的身份锚点
- 被简化掉的照片细节
- 需要你确认的地方
- 第一张正面全身图提示词
Generate or write a Chinese prompt for creating a front-view full-body confirmation image in the target comic style. The first confirmation image must be a complete full body, solid-color background, and pure front view; outputting a headshot, half body, or mostly full body is not allowed.
After the user confirms the front-view full-body confirmation image, then generate or write the three-view prompt. The three-view must first be confirmed by the user before entering the expression-action pack stage.
After the user confirms the three-view, then generate or write prompts for 8 single-action expression reference images; if there is an accompanying character, then also make a three-view for the accompanying character.
When generating a single expression-action reference image, the confirmed three-view.png must be used as a reference image input; each of the 8 images must again attach the same three-view as a reference. It cannot be referenced only once at the start of the expression pack stage. If the three-view cannot be used as a reference image, stop advancing and ask the user to upload or confirm the three-view.
The expression-action pack must be generated as 8 separate images according to 01 to 08 . Do not generate one large image containing 8 expressions, do not generate a contact sheet, nine-grid, collage, sticker collection, and do not split afterward.
Before accepting images, a final quality check must be performed:

Whether the identity anchor is retained
Whether the three-view is used as the reference image for the expression-action images
Whether the expression-action pack has exactly 8 images
Whether each image is an independent single image, rather than split from a collection
Whether the background color, character proportions, lines, and color scheme of the 8 images are consistent
Whether the eye and face rules are followed
Whether the clothing and accessories are stable
Whether the body structure is normal
Whether features of other IPs are mistakenly borrowed
Whether the final image comes from an image generation or image editing route

After completing each stage, proactively ask the user for the next step; do not wait for the user to say "continue" on their own.
Only after user confirmation, save the qualified files into the archive folder, or clearly tell the user which archive package should be installed/delivered.

Expression-action pack consistency gate

Before generating the expression-action pack, the following conditions must be met:

There is already a user-confirmed three-view.png。
In the current conversation or task, this three-view can be used as an image reference input.
Each expression-action prompt explicitly requires referencing the three-view and maintaining overall character consistency.
8 independent single images have been clearly planned, and it has been confirmed that no collection will be generated and then split.

If any condition is not met, do not generate the expression action pack. Instead, proactively ask:

下一步我建议：先上传或确认三视图，再生成表情动作包，这样能保证角色一致。
请你确认：现在是否使用这张三视图作为后续表情动作图的参考？
可选方向：
1. 使用当前三视图继续生成表情动作包
2. 先重新生成或修正三视图
3. 暂停表情包，先补充更多角色参考

The prompt for the expression action image must include this sentence:

请以上传的三视图作为同一个 IP 的角色参考，严格保持头身比例、整体轮廓、脸部特征、眼睛规则、服装、配饰、颜色锚点和线条风格一致；只改变本张图的表情、动作和道具。

The expression action pack must be generated using 8 independent prompts, with prompt numbers 01 to 08. Each prompt must clearly state:

This is a single standalone image, not one panel of a collection.
The background color must match the other 7 images.
The character proportions, silhouette, clothing, accessories, line thickness, and color scheme must match the three-view.
Each generation must actually use the same confirmed three-view as the image reference input.
Only the emotion, action, and optional props of this image may be changed.
No contact sheets, nine-grid layouts, collage images, sticker collections, storyboard pages, or any large image requiring later splitting.

IP Extraction Rules

Extract stable identity anchors: silhouette, hair/head shape, facial rhythm, signature accessories, clothing logic, body proportions, and color anchors.
Translate source details into a simple flat comic grammar. Avoid photographic texture, realistic lighting, complex fabric textures, skin pores, and fine hair strands.
Use concrete drawing rules. Do not just write abstract words like "cute," "smart," or "premium" unless followed by a visible visual decision.
If a feature is unclear or ambiguous, mark it as 未知, and provide 1-2 conservative options for the user to confirm.
If the lower body, shoes, tail, bottom structure, or body proportions are not visible in the source image, mark them as 根据现有信息保守推断, and do not treat them as confirmed facts.
Separate style DNA from IP DNA: shared styles such as lines, paper texture, and composition belong to the downstream illustration skill; character-unique identity goes into IP_DNA.md。
If an existing IP profile was referenced, only reuse the document structure and QA strictness. Do not copy hair, clothing, earrings, pets, props, or color schemes unless the new IP itself has these features.

Confirmation Criteria

Do not treat the first generated character as the final version. The first image is only a confirmation probe.

The user must at least be able to answer:

Does this still look like the person, mascot, or toy originally intended?
Is this confirmation image already a complete full body, rather than a half body or three-quarter body?
Which 2-3 details are wrong?
Which detail must absolutely never drift in the future?
Does this character need a companion character?
Is this output style suitable for future article illustrations?
Does the expression action pack clearly follow the three-view from the previous step, rather than looking like a redesigned character?
Is the expression action pack exactly 8 independent standalone images, with background colors and characters consistent?

If the user corrects any identity anchor, you must first update IP_DNA.md, then continue generating more views.

Downstream Handoff

After the IP profile is confirmed, provide a brief handoff message:

当前 IP 档案: <path/to/IP_DNA.md>
必需参考图:
- <front-full-body.png>
- <three-view.png>
- <expression folder>
提示词胶囊:
<从 IP_DNA.md 提取的一段中文提示词>
硬性质检:
<5-8 条不可妥协的检查项>

The downstream comic illustration skill should read this profile package, and should not rewrite the character rules in its own main instructions.

After completing the handoff, you must continue asking the user:

下一步我建议：把这个 IP 档案接入一个下游漫画信息图技能，试跑一篇文章。
请你确认：现在是要接入下游技能，还是继续为另一个 IP 建档？
可选方向：
1. 接入下游漫画信息图技能并试跑一篇文章
2. 继续优化当前 IP 的表情动作包
3. 为另一个用户创建新 IP 档案