Brclio/Skills

Content Illustration Character IP Standard Pack

Goal

Turn a personal, creator account, or brand description into a reusable content-illustration character. The final delivery includes character values, visual DNA, front view, three views, character reference board, layered prompts, and consistency test images.

Core Rules

• Designed for non-designers: prioritize asking users to upload images they like or existing assets; use minimal terminology and offer 2-3 clear options.
• Can start without any images, descriptions, or if the user answers "I don't know"; only ask questions that are critical for design, and limit to 3 per round.
• Only generate images after the user confirms the character DNA; separate aesthetic confirmation from visual specification acceptance.
• The standard front view is the only authoritative original asset; subsequent views must not be redrawn to replace it.
• For three views, series images, and sticker sets, always use the same standard front view or character reference board as a reference.
• Generate and accept each image independently; do not generate a grid and then crop.
• Use relative paths for the official Pack; do not bind to local paths.

1. Quick Information Gathering

Start with:

Before we begin, you can choose any of the following: upload 1-5 images you like or dislike; paste 1-3 pieces of representative content; describe yourself or your brand briefly; or say "I don't know" and I'll recommend options.

Choose one or mix:

1. Upload Images: 1-5 IP, illustration, or toy images you like or dislike; for each, note "like," "dislike," or "only like the colors/proportions/texture."
2. Provide Existing Content: Paste 1-3 representative articles, posts, account bio, or brand introduction for the agent to extract values, tone, and audience.
3. Simple Description: Explain who you or your brand are, what content you create, what feeling you want to convey, and where the character will be used.
4. Don't Know at All: The agent recommends 2-3 directions based on content theme and audience.

Images are processed in two categories:

• Inspiration References: Extract only shape, proportions, facial features, color palette, texture, detail level, and feel; do not replicate specific characters, clothing, logos, props, or unique compositions.
• Own Assets: Personal photos, existing characters, logos, product images, or brand colors can serve as identity or brand constraints; first confirm which elements must be preserved.

After reading images, summarize the user's liked and disliked visual features in one sentence for confirmation; do not ask the user to articulate design terminology.

Also extract from images or descriptions: identity, content theme, audience, primary uses, character contradictions, personal values, expression boundaries, and brand constraints. When information is insufficient, only ask questions that would change character shape, value stance, or usage.

2. Provide Character Directions

Offer 2-3 directions, each containing only:

• Character shape and silhouette
• Personality and value stance
• Signature recognition point
• Suitable content
• One main risk

Only build the DNA after the user selects a direction.

3. Confirm Character DNA

meta:
  name:
  mode: creator | brand
  audience:
  primary_uses: []

personality_values:
  one_line_core:
  soul_tension:
  core_values: []           # maximum 3
  core_belief:
  content_stance:
  behavior_patterns: []
  voice:
  boundaries: []

visual_dna:
  form: character | object
  silhouette:
  body_proportions:
  face_lock:
  signature_props: []       # 1-2
  accent_rule:
  palette:
    primary: {name: "", hex: ""}
    secondary: {name: "", hex: ""}
    accent: {name: "", hex: ""}
    outline: {name: "", hex: ""}
    background: {name: "", hex: ""}
  movable_parts: []
  forbidden_drift: []
  content_actor_notes:

Values must guide how the character expresses content; do not use vague words like "gentle, free, brave" alone.

4. Confirm Art Style

Before generating any image, the user must confirm the art style; do not decide silently based on reference images.

Provide four short style cards:

1. Healing Illustrated Hand-drawn: Like independent picture books and hand-drawn creative characters, warm with paint and paper texture; suitable for emotional and lifestyle content, details tend to drift more easily.
2. Minimalist Symbolic Cartoon: Use the "few lines, few colors, high recognition" logic of Miffy or LINE FRIENDS to explain; suitable for avatars, stickers, and light branding, weaker at complex actions.
3. Blind Box / Designer Toy: Use the "big head proportion, rounded shapes, vinyl material, display feel" of Pop Mart-style designer toys to explain; suitable for brand characters, higher demands on three views.
4. Everyday Flat Comic: Use the logic of Snoopy or Moomin comics where "characters can consistently act and enter daily scenes" to explain; suitable for long-term content illustration.

Famous IPs are only used as text-based cognitive anchors; do not write their names into prompts or require replication of their character appearance or drawing style.

If providing visual style cards, all four cards must use the same original example character, same pose, and same color palette, changing only the drawing method and material so the user can compare real style differences. Each card only shows: style name, original example image, suitable content, and main risk.

• If the user has no reference images: recommend 1 style based on use case, with 1 backup option.
• If the user uploaded reference images: first explain which style the images are closest to, then recommend 1-2.
• Users can directly select a number or reply "Go with your recommendation."

After the user explicitly confirms, write style_id, material, and corresponding style_lock into the Visual DNA. Do not generate the standard front view before confirmation.

Read style_refs/STYLE-LOCKS.md. Reference images are only used to convey material, lines, and rendering; do not replicate the characters, color palettes, clothing, props, or compositions from them.

5. Generate Standard Front View

Standard front view requirements:

• Front-facing, eye-level, neutral standing or neutral floating pose
• Full body, limbs/components, and signature items clearly visible
• Light solid background, no scene, text, or obstructions
• No exaggerated poses to hide body proportions

After the user confirms they like it and it passes the consistency check, save the original image as character-front.png.

6. Generate Three Views and Character Reference Board

Using the standard front view as a reference, generate:

• Side view
• Back view
• Close-up of the face
• Close-up of signature items
• Standard color swatch

Use the original standard front view directly for the front image. All views must maintain the same proportions, face, components, signature items, intrinsic colors, and material.

Arrange accepted original views losslessly into character-reference-board.png. Do not redraw or alter the standard front view during layout. For subsequent cross-agent generation, prioritize uploading this character reference board.

7. Generate Reusable Prompts

Do not write fixed poses, compositions, or backgrounds into the character locks.

identity_lock: ""          # body, proportions, face, signature items, intrinsic colors
style_lock: ""             # material, lines, rendering method
values_tone: ""            # use when content requires it
pose_slot: ""              # replace per image
environment_slot: ""       # replace per image, can be empty
composition_slot: ""       # replace per image
output_spec: ""            # aspect ratio, size, background requirements
negative_lock: ""          # permanently forbidden identity and style drift

Also deliver:

• standard_front_prompt
• scene_prompt_template
• If the user needs sticker sets, also deliver sticker_prompt_template

Scene prompts are concatenated in this order:

identity_lock + style_lock + values_tone
+ pose_slot + environment_slot + composition_slot
+ output_spec + negative_lock

8. Consistency Acceptance

Using the character reference board and scene template, independently generate three test images:

1. Common action
2. Large action
3. Light scene content illustration

Check each image out of 100:

Item	Score
Silhouette and shape	20
Body proportions and components	15
Face lock	20
Signature items	15
Intrinsic and accent colors	10
Material and art style	10
Character feel	5
Output specification	5

Official images must score 90/100. Species change, incorrect number of limbs/components, face structure change, missing signature items, or redrawing of the standard front view are automatic failures.

When failing, only fix the failing DNA, prompt, or view; do not regenerate the entire set randomly.

9. Deliver Standard Pack

<ip-name>-standard-pack-v1/
├── IP-PACK.md
├── manifest.yaml
├── references/
│   ├── character-front.png
│   ├── view-side.png
│   ├── view-back.png
│   └── character-reference-board.png
├── prompts/
│   ├── prompt-locks.yaml
│   └── copy-paste-prompts.md
└── tests/
    ├── test-01-common-action.png
    ├── test-02-large-action.png
    ├── test-03-light-scene.png
    └── identity-qa.md

manifest.yaml records the Pack version, used agent/model, reference image method, relative paths to official assets, and known limitations. The status can only be set to locked after all three views and three test images have passed.

Optional Sticker Set

Only generate after the user explicitly requests. Generate each sticker independently with the same character reference board uploaded as reference; do not allow drift in intrinsic colors or signature items. Check for transparent background, dimensions, safe margins, and thumbnail readability according to the target platform; do not force a fixed quantity.

Final Checklist

• Values guide how the character expresses content
• DNA includes proportions, face, signature items, hex color swatch, and forbidden drift
• User explicitly confirmed art style, Visual DNA includes style and material
• Standard front view is the confirmed original front image
• Three views and character reference board are complete
• Layered prompts have no fixed-pose conflicts
• All three test images score 90
• Pack uses only relative paths
• Brand character does not replicate features of specific brands or protected characters