dreamer-star-ip

English · 中文

⭐ dreamer-star-ip

Turn one idea into one memorable editorial illustration.

A reusable AI-agent skill built around 小星妍: a honey-yellow star character, generous negative space, structural hatching, and a clear visual contrast.

Overview

dreamer-star-ip turns Chinese articles, ideas, and arguments into minimalist conceptual illustrations. Instead of decorating a page with a generic image, it finds a cognitive anchor and expresses it through one focused visual metaphor.

The recurring character is 小星妍 (Dreamer Yanyan): a slightly tilted, honey-yellow five-pointed star with a small curved Dreamer wordmark, hand-drawn hatching, and a contextual status glyph. She always performs the central action; she is never a mascot placed in the corner.

Visual formula

One image = one idea + 55–65% negative space + three hatching areas + one clear contrast hook

Best for

• Editorial illustrations for Chinese articles and newsletters
• Xiaohongshu posts and social media covers
• Presentation and Notion document visuals
• Abstract ideas that benefit from a concrete visual metaphor

Not designed for

• Portrait-led or photo-based covers
• Polished 3D commercial illustration
• Dense infographics, architecture diagrams, or presentation dashboards

What You Get

Output	What it contains
Shot list	4–8 illustration opportunities tied to specific passages and cognitive anchors
Production prompt	A structured 12-block English prompt covering composition, action, labels, color, and anti-slop rules
Final image or copy-ready prompt	The image is generated when the host agent has an image tool; otherwise the complete prompt is returned
Placement note	A short explanation of where the image belongs and how its metaphor supports the argument

Gallery

Quick Start

1. Clone

git clone https://github.com/yanliudesign/dreamer-star-ip.git
cd dreamer-star-ip

2. Install for your agent

The repository root is the skill directory. The commands below create symbolic links, so a later git pull also updates the installed skill.

Agent	Personal skill directory	Invoke with
Claude Code	~/.claude/skills/	/dreamer-star-ip
Codex	$CODEX_HOME/skills/ (default: ~/.codex/skills/)	$dreamer-star-ip
OpenCode	~/.config/opencode/skills/	Ask it to use dreamer-star-ip
OpenClaw	$OPENCLAW_STATE_DIR/skills/ (default: ~/.openclaw/skills/)	/dreamer-star-ip
Hermes Agent	~/.hermes/skills/	/dreamer-star-ip

mkdir -p "$HOME/.claude/skills"
ln -s "$(pwd)" "$HOME/.claude/skills/dreamer-star-ip"

CODEX_HOME="${CODEX_HOME:-$HOME/.codex}"
mkdir -p "$CODEX_HOME/skills"
ln -s "$(pwd)" "$CODEX_HOME/skills/dreamer-star-ip"

mkdir -p "$HOME/.config/opencode/skills"
ln -s "$(pwd)" "$HOME/.config/opencode/skills/dreamer-star-ip"

OPENCLAW_STATE_DIR="${OPENCLAW_STATE_DIR:-$HOME/.openclaw}"
mkdir -p "$OPENCLAW_STATE_DIR/skills"
ln -s "$(pwd)" "$OPENCLAW_STATE_DIR/skills/dreamer-star-ip"

mkdir -p "$HOME/.hermes/skills"
ln -s "$(pwd)" "$HOME/.hermes/skills/dreamer-star-ip"

Restart the agent or begin a new session after installation. For project-only use, install the skill inside that project under .claude/skills/ (Claude Code), .opencode/skills/ (OpenCode), or skills/ (OpenClaw).

3. Run your first workflow

The examples use Codex syntax. Replace $dreamer-star-ip with /dreamer-star-ip in Claude Code, OpenClaw, or Hermes. In OpenCode, ask the agent to use dreamer-star-ip.

Plan a series before generating

Use $dreamer-star-ip to analyze the article below. Do not generate images yet.
Create a shot list of about five illustrations. For each one, include placement,
cognitive anchor, visual metaphor, contrast hook, character action, status glyph,
and labels.

<paste article>

Generate a complete illustration series

Use $dreamer-star-ip to create four 3:4 editorial illustrations for the article
below. Give every image a distinct metaphor, generate each image separately,
run the QA checklist, and include a placement note and reproducible prompt.

<paste article>

Illustrate one idea

Use $dreamer-star-ip to illustrate this idea:
"Trust is built one piece of evidence at a time."
Choose a fitting action from the action library and do not reuse an example
composition.

Image generation depends on the tools available to the agent. If no image tool is available, the skill returns the full prompt instead of claiming that an image was generated.

How It Works

The skill follows a five-stage workflow:

1. Extract the cognitive anchor — identify the judgment, transition, state, process, or metaphor worth illustrating.
2. Create a shot list — define placement, concept, contrast, character action, status glyph, and short labels before generating.
3. Build the prompt — use a 12-block prompt structure with explicit format, composition, color, hatching, and anti-slop constraints.
4. Run QA — evaluate character geometry, participation, negative space, hatching, labels, background, and visual noise.
5. Deliver reproducibly — return the image when possible, the complete prompt, and a placement note.

The primary format is 3:4 portrait. The system also includes composition rules for 4:3 landscape and 1:1 square.

The Design System

This repository extends a general editorial-illustration workflow into a repeatable character-IP system:

• Character specification — geometry, personality, poses, signatures, companion relationships, and failure modes
• 16 status glyphs in five intent families — a second-read clue showing what the character is thinking or feeling
• 48 reusable actions in five categories — star-specific metaphors, general poses, narrative scenes, reaction foundations, and workplace actions
• Quantified composition rules — format, negative-space ratio, contrast axis, label limits, and hatching placement
• 12-block prompt structure — consistent identity across topics and image models
• P0/P1/P2 quality control — clear guidance for regenerating, editing locally, or accepting a minor variation

Core principles

• One image, one idea. Never turn the illustration into an infographic.
• Keep 55–65% negative space. Empty space gives the concept room to breathe.
• Use exactly three structural hatching areas. Hatching elsewhere becomes visual noise.
• Create a one-second contrast hook. Examples include top versus bottom, before versus after, definition versus optimization, or individual versus group.
• Treat the status glyph as the second-read clue. It should explain the character's primary thought or feeling without dominating the scene.
• Make the character act. A small jump is better than a static pose; 小星妍 must never appear as decoration.

Use It as a Prompt Library

You do not need to install the repository as a skill. You can copy a complete prompt from examples/ai-era-product-interview-v6.2.md into GPT Image, Nano Banana, Midjourney, or another image model.

To build a new scene:

1. Browse action-library-preview.html or read references/action-library.md.
2. Choose one of the 48 actions based on the topic.
3. Adapt the action inside the Single visual concept and Composition sections of references/prompt-template.md.
4. Add no more than two short Chinese labels.
5. Review the output with references/qa-checklist.md.

Repository Guide

dreamer-star-ip/
├── SKILL.md                        Main entry and five-stage workflow
├── references/
│   ├── style-dna.md                Color, negative space, and hatching
│   ├── xiaoxingyan-ip.md           Character specification and pose library
│   ├── status-glyphs.md            16 status glyphs in five families
│   ├── action-library.md           48 reusable actions in five categories
│   ├── composition-patterns.md     Contrast hooks and format rules
│   ├── prompt-template.md          12-block English prompt template
│   └── qa-checklist.md             QA levels and anti-slop checklist
├── example/                        Nine finished gallery images
├── action-library-preview.html     Visual action-library browser
├── THIRD_PARTY_NOTICES.md          Attribution and third-party notices
└── examples/
    ├── ai-era-product-interview-v6.2.md   Seven complete v6.2 prompts
    ├── six-vertical-illustrations.md      Six earlier portrait prompts
    ├── four-poses/                        Pose studies
    └── ip-manual/                         Character manual

FAQ

The agent cannot find dreamer-star-ip

1. Confirm that the installed path points to this repository:

   ls -l "$HOME/.claude/skills/dreamer-star-ip"
   ls -l "${CODEX_HOME:-$HOME/.codex}/skills/dreamer-star-ip"
   ls -l "$HOME/.config/opencode/skills/dreamer-star-ip"
   

2. Confirm that SKILL.md exists at the root of the linked directory.

3. Restart the agent or open a new session; skill discovery usually happens when a session starts.

4. Use the platform-specific invocation syntax from the installation table.

ln -s says File exists

An installation already exists at that destination. Inspect it with ls -l <path> before changing anything. If it already points to this clone, there is nothing to do. Otherwise, remove only that specific old link and run the installation command again.

How do I update the skill?

cd /path/to/dreamer-star-ip
git pull

You do not need to recreate the symbolic link. Start a new agent session so the updated instructions are loaded.

How do I uninstall it?

Remove only the platform-specific symbolic link, not the cloned repository:

rm "$HOME/.claude/skills/dreamer-star-ip"

Replace the path with the appropriate Codex, OpenCode, OpenClaw, or Hermes destination when needed.

Why did I receive a prompt but no image?

The skill defines an illustration workflow but does not include an image model. When the current agent has no image-generation tool, it intentionally returns a complete prompt. Paste that prompt into GPT Image, Nano Banana, Midjourney, or another supported model.

The Chinese labels are misspelled or garbled

Image models remain inconsistent at rendering Chinese text. Use one or two short, common labels. If the illustration is otherwise correct, regenerate or inpaint only the label area. When accuracy is critical, generate the artwork without labels and add them later in a design tool.

The result looks too cute, busy, or like an infographic

Run references/qa-checklist.md again. Keep 55–65% negative space, exactly three controlled hatching areas, no decorative sparkles, no large title or legend, and no more than two labels. 小星妍 must perform the central action instead of standing beside it.

Should I regenerate or edit the current image?

• Regenerate: incorrect star geometry, honey-yellow color leaking into other elements, wrong background, decorative companion stars, or an oversized status glyph
• Edit locally: incorrect label text, misplaced hatching, or a slightly stiff pose
• Usually acceptable: mildly reduced negative space or an imperfect status glyph that still has the correct color and placement

Acknowledgements

This project draws on Ian's open-source Ian Xiaohei Illustrations, especially its approach to extracting cognitive anchors from Chinese writing, building shot lists, and structuring editorial-illustration workflows.

dreamer-star-ip does not replace Ian's broader illustration toolkit. It extends the method into a reusable character-IP system focused on consistency across a series.

See THIRD_PARTY_NOTICES.md for attribution and third-party license details.

License

MIT. See LICENSE.

The dreamer-star-ip character, status-glyph system, and composition language are open for learning and remixing. Please provide attribution when using them commercially.