personal-ip-builder

Personal IP Builder — From Reference Image to Digital Assets and Tutorial Illustration Skill

The goal is not to "draw one good-looking avatar", but to deliver a complete set of digital assets that can support self-media creation, and optionally distill the assets into a repo-only tutorial illustration skill.

When to Use

The user has a favorite anime/character reference image and wants to turn it intotheir ownoriginal personal IP
Needs a complete set: character sheet, avatar, full-body character art, expressions and poses, content scenes, stickers, not just a single image
Once assets are stable, wants to make a skill that "produces tutorial illustrations from just one sentence"

Two Layers of Delivery

Layer	Delivery	Stop Point
First layer (default)	The complete digital asset set + at the asset root directory `character-contract.md`: character sheet, visual specification board, avatar, full-body character art, expressions and poses, content scenes, stickers, horizontal key visual	If the user only wants assets, stop here
Second layer (optional)	Based on confirmed assets, create the repo-only tutorial illustration skill; the skill referencesin-repo assets and contracts, without copying PNGs into the skill directory	Only when assets are stable and the user explicitly requests it

Core Principles

The input determines who it is; this skill determines how to make it a reusable asset. The reference image only provides mood, temperament, and direction; copying the original character's face, clothing, composition, or logo is forbidden.
Define identity and contract first, then generate images in batch. Do not batch-generate before the "character design contract" is confirmed.
Produce the character sheet first, then expand to the complete asset set. If the character sheet does not pass, do not proceed to batch avatar/expression/sticker generation.
The asset pack is one endpoint; the tutorial illustration skill is an optional second endpoint. If the user only wants assets, that is also a valid stopping point.
One question at a time; one core per image; white background with dark gray line art; the character must be doing something。

Stage Overview

Check workspace → read reference image → grill-me design interview → output contract and wait for confirmation → character sheet first, then expand the complete asset set → review and targeted fixes → optional tutorial illustration skill

Stage 0: Check Workspace

Confirm the current directory is the root of the user's project or notes repository, and check:

Whether there is already a .agents/skills/ or equivalent skills directory
Whether there is already an assets directory such as 03-Assets/ or a user-specified assets path
Whether grill-meis available; if not, prompt to install:npx skills@latest add mattpocock/skills, interactively select only grill-me, current Project/repo-only, not global -g
Whether the current environment hasimage generation capability that supports reference images; if not, explain the gap, deliver only the contract and prompts, and do not pretend images have been generated

Stage 1: Read Reference Image

Review the reference images uploaded by the user. Reference images only provide emotion and direction; replicating the original character is prohibited.

Can extract: emotion, sense of proportion, color memory points, sense of motion
Must be made original: facial detail, outfit combinations, professional identity, brand accessories, fixed symbols
How to connect with the user's real content positioning

Do not draw by copying; turn "I like this character" into "which visual languages do I want to borrow".

Stage 2: Design interview (prefer grill-me)

Ask only one question at a time; give a recommended answer and reasoning for each question; positioning first, then appearance. For question order see references/design-interview.md。

At minimum confirm:

Content positioning and target audience
Anime-style avatar or an independent fictional character
Gender expression, personality keywords (prefer words that can drive the visuals, such as "curious action-taker")
Art style, head-to-body ratio, hairstyle, eyes
Fixed outfit, primary/secondary colors, forbidden colors
One stable identifying accessory
One brand letter or visual symbol
The most common core action in the content
At least 4 expressions or states corresponding to the content production workflow
Uses: avatar / tutorial / article / cover / sticker, etc.
First-version asset list

Continuously maintain a temporary "character design contract" so confirmed items are not lost between rounds.

Stage 3: Output the character design contract

After the interview ends, summarize in a concise and complete format and ask the user to confirm. Do not generate images before confirmation. The contract must contain at least:

Name and IP positioning
Personality and way of speaking
Hard appearance constraints
Fixed accessories and brand identity
Color and line art specifications
Common actions and expressions
Usage scenarios and first-version asset list
Originality boundaries and prohibited items

After confirmation, write the same contract into the asset root directory's character-contract.md。

Stage 4: Generate the full set of digital assets

After the contract is confirmed, the Agent automatically organizes prompts and generates images, do not make the user hand-write long prompts.

Default visuals and composition

Default landscape 16:9; use 1:1, 4:5, 5:2, etc. only when the user explicitly requests it
Pure white background, dark gray hand-drawn line art, clean flat coloring, a small amount of soft shadow
Primary color, hairstyle, accessories, outfit, and brand identity consistent throughout
The character must be the core action subject of the image, must be doing something, and cannot just stand there as decoration
Single image, single core meaning, do not combine multiple scenes into one multi-panel large image
By default, do not generate text; only 1–5 short Chinese words explicitly provided by the user may appear in the image
Prohibit extra characters, third-party logos, watermarks, long paragraphs, complex PPTs, and extra limbs

Asset list that must be delivered (full set)

Generate in order, with each category output as a separate image. For acceptance items see references/asset-checklist.md。

Character setting sheet: full-body main pose + key identifying points + core expression thumbnails, used to verify consistency
Visual specification sheet: a one-page specification of colors, line art, accessories, and prohibited items
Avatar: front view or half-body, suitable for account avatars and small-size cards
Full-body standing illustration: suitable for tutorials and cover action expansion
Expression and action group: 4–6 of them, recommended to align with the content flow, e.g., discovery/focus/failure/happy sharing/confusion/thinking
Content scene group: the character is operating the interface, troubleshooting, explaining the process, etc., to help the model understand "how the character participates in the tutorial"
Stickers and reaction images: suitable for social media and emotional moments in the text
Horizontal key visual: usable for article or long-form covers

Before the setting board is approved, do not batch-generate 3–8.

Archive directory

Save to the user-specified path, default:

03-Assets/YYYY-MM-DD-<主题>-personal-ip/
├── character-contract.md
├── 01-角色设定/
├── 02-头像与立绘/
├── 03-表情与动作/
├── 04-内容场景/
└── 05-贴纸与装饰/

Do not overwrite old versions; for revisions use -v2、-v3. Keep reusable setting boards and visual specification boards. Put the horizontal key visual into 01-角色设定/ or the user-specified cover directory, and specify the path in the contract.

Stage 5: Review, inspection, and targeted fixes

After each image is generated, inspect it before showing it to anyone:

Whether it is still the same person: hairstyle, accessories, clothing, color scheme, brand symbols
Whether the character is actually doing something and whether the action is readable
Whether the silhouette and accessories are still distinguishable at small sizes
Whether it is too crowded like a PPT; whether it accidentally replicates the reference image
Whether there are extra limbs, severed limbs, or garbled text
Whether it still has a white background, dark gray line art, and a single core per image

When a problem occurs, change only one main cause, e.g., "only remove the third hand; the rest of the composition and title stay unchanged". Do not tear down and redo the whole set. If it is still stuck on a key defect after two rounds, deliver the current best and explain the remaining issues.

Stage 6 (optional): Create a tutorial illustration Skill

Once the assets are stable and the user requests it, create a repo-only tutorial illustration skill. Do not copy other IPs' characters or case compositions; transfer only the "explanatory illustration method":

Can produce an image even when the user gives only one sentence of tutorial/process/tool/concept
Does not require reading the full text
Default 16:9, single core per image
The character must participate in the core action
White background, dark gray line art, user IP color scheme
A small amount of Chinese annotations provided by the user
Split multiple core meanings into multiple images; do not combine them into multiple panels
After generation, check consistency, composition, text, and limbs
Save images to 03-Assets/YYYY-MM-DD-<主题>-ip-illustrations/
Place the skill in .agents/skills/<skill-name>/, containing at least SKILL.md; if there is UI metadata, include agents/openai.yaml
skill Referencethe paths of assets in the repository and character-contract.md . Do not copy PNGs into the skill directory, which would create duplicate versions

Before creating, check whether a similar skill already exists. After creating, run the validation scripts available in the environment and tell the user the results.

The tutorial illustration skill's usage examples should be written into that skill's README, for example:

使用 $<tutorial-skill-name>，画一张 16:9 教程配图：
角色正在排查 API 失败，一手扶配件，另一手指向被标出的错误参数。纯白背景，大量留白，不要长文字。

Deliverables list

Default deliverables:

Confirmed character-contract.md
Complete digital asset directory (setting boards, specification boards, avatars, character art, expressions, scenes, stickers, key visuals)
Brief note: which undisplayed parts the design filled in; which images require the user's final aesthetic review
If the user requests it: a repo-only tutorial illustration skill + the result of one real-topic trial run

Boundaries and prohibitions

Do not batch-generate images before the contract is confirmed
Do not treat reference images as replication templates; take only emotion and direction
Don't inflate “one good-looking image” into “one magic prompt”
Don't skip the character consistency check
Don't copy others' IP, logos, or case compositions into assets or skills
Don't commit users' personal PNG asset packs into this skill repository
When there is no image generation capability, deliver only the contract and prompts; don't pretend images have been generated

References

references/asset-checklist.md — complete digital asset acceptance checklist
references/design-interview.md — order of design interview questions
Long-form source on the method:https://x.com/chasen_liao/status/2079467790252294636