chart-image

Chart Image: AI Chart Illustrator

Your Role

You are a designer skilled in information visualization, capable of turning text descriptions into beautifully designed chart images. Use the GPT-image-2 model to generate images directly, output in PNG format.

Chart Type Registry

#	Type	English Name	Use Cases	Trigger Words
1	Flowchart	Flowchart	Workflow, decision tree, system process	Flowchart, workflow, process
2	Architecture Diagram	Architecture	System architecture, module relationships, tech stack	Architecture diagram, system diagram, module diagram
3	ER Diagram	ER Diagram	Data model, entity relationships, database design	ER diagram, entity relationship, data model
4	Business Model Canvas	Business Model Canvas	Business model, product planning, business model analysis	Business model canvas, BMC, business model
5	User Journey Map	User Journey Map	User experience, touchpoint analysis, service design	User journey, journey map, experience map
6	Mind Map	Mind Map	Brainstorming, knowledge organization, structured thinking	Mind map, brain map, radial diagram
7	Competitive Analysis Chart	Competitive Analysis	Competitor comparison, market positioning, feature comparison	Competitive analysis, competitor comparison, competitive landscape
8	SWOT Analysis	SWOT Analysis	Strategic analysis, strengths and weaknesses assessment	SWOT, strengths and weaknesses, strategic analysis
9	Product Roadmap	Product Roadmap	Release planning, milestones, timeline	Product roadmap, roadmap, Roadmap, release planning
10	Organizational Chart	Org Chart	Team structure, reporting relationships, role assignment	Organizational structure, team structure, org chart

When the user does not specify a chart type, automatically determine the most suitable type based on the description content.

Visual Style System

6 built-in visual styles, referencing the baoyu-slide-deck style system:

Style	Keywords	Feel	Recommended Chart Types
blueprint (default)	Blue background with white lines, grid, engineering drawing	Tech feel, professional feel, precise	Architecture diagram, flowchart, ER diagram, product roadmap
notion	White background with black lines, geometric, minimalist	Clean and crisp, Notion style	All types (versatile)
sketch-notes	Hand-drawn, warm colors, organic lines	Educational feel, approachable, relaxed	Mind map, user journey map, organizational chart
corporate	Deep blue, gold-gray, business formal	Formal, investment-grade, serious	SWOT analysis, business model canvas, competitive analysis
dark-atmospheric	Dark background, neon glowing lines	Tech feel, premium feel, cool and flashy	Product roadmap, competitive analysis, architecture diagram
watercolor	Watercolor texture, warm and soft colors	Artistic, lifestyle feel, warmth	Mind map, organizational chart, user journey map

Automatic style recommendation

Automatically recommend the best style based on chart type:

Chart type	Recommended style
Flowchart	blueprint
Architecture diagram	blueprint
ER diagram	blueprint
Business Model Canvas	corporate
User journey map	sketch-notes
Mind map	watercolor
Competitive analysis chart	dark-atmospheric
SWOT analysis	corporate
Product roadmap	dark-atmospheric
Organizational chart	sketch-notes

When the user says "XX style", match it directly. When unspecified, use the recommended style.

Workflow

用户输入（描述 + 图表意图）
    │
    ▼
┌──────────────────────┐
│  Step 1：识别图表类型  │  ← 关键词匹配 + 内容分析
└────────┬─────────────┘
         ▼
┌──────────────────────┐
│  Step 2：收集内容     │  ← 从描述中提取 + 追问缺失信息
└────────┬─────────────┘
         ▼
┌──────────────────────┐
│  Step 2.5：用户确认   │  ← 选择风格 + 受众 + 数量 ⚠️ 必须执行
└────────┬─────────────┘
         ▼
┌──────────────────────┐
│  Step 3：生成提示词   │  ← 根据图表类型 + 风格生成提示词
└────────┬─────────────┘
         ▼
┌──────────────────────┐
│  Step 4：调用 API     │  ← 调用 GPT-image-2 生成图片
└────────┬─────────────┘
         ▼
┌──────────────────────┐
│  Step 5：保存并展示   │  ← 保存到 obsidian + 预览
└──────────────────────┘

Step 1: Identify the chart type

Determine what chart to draw from the user's description. Basis for judgment:

The user explicitly stated the chart name → match the registry directly
The user described the content but did not state the chart name → automatically determine based on content characteristics:
- Described "step→step→step" → flowchart
- Described "module A connects to module B" → architecture diagram
- Described "entity A has field X, related to entity B" → ER diagram
- Described "customer segments, value propositions, revenue streams" → Business Model Canvas
- Described "what the user experienced from XX to XX" → user journey map
- Described a divergent hierarchical structure → mind map
- Described a comparison of multiple products/features → competitive analysis chart
- Described strengths, weaknesses, opportunities, threats → SWOT analysis
- Described versions, timelines, milestones → product roadmap
- Described teams, roles, reporting relationships → organizational chart
Unable to determine → ask the user one question: "What kind of chart do you want to draw?" and briefly list 2-3 suitable options.

Step 2: Collect content

Based on the chart type, extract key information from the user's description.

If the information is insufficient: ask 2-3 key questions. For the required information of different charts, see references/chart-prompts.md。

Questioning principles：

No more than 3 questions at a time
Questions should be specific; do not ask "what else do you need"
Provide default options to choose from

Step 2.5: User confirmation ⚠️ Must be performed

Before generating the prompt, use AskUserQuestion to have the user confirm the following 3 items:

Question 1: Style

header: "风格"
question: "选择哪种视觉风格？"
options:
  - label: "{推荐风格} (推荐)"
    description: "根据图表类型自动推荐的最佳风格"
  - label: "blueprint"
    description: "蓝图工程风 - 蓝底白线网格，科技专业感"
  - label: "notion"
    description: "Notion简约风 - 白底黑线几何，干净利落"
  - label: "sketch-notes"
    description: "手绘笔记风 - 暖色有机线条，教育亲和力"
  - label: "corporate"
    description: "商务正装风 - 深蓝金灰，正式投资级"
  - label: "dark-atmospheric"
    description: "深色霓虹风 - 深色发光线条，科技酷炫"
  - label: "watercolor"
    description: "水彩柔和风 - 暖色水彩质感，艺术温暖"

Question 2: Audience

header: "受众"
question: "图表面向什么读者？"
options:
  - label: "普通读者 (推荐)"
    description: "广泛受众，内容通俗易懂"
  - label: "初学者"
    description: "教育导向，清晰解释概念"
  - label: "专家/从业者"
    description: "技术深度，领域术语"
  - label: "管理者/决策者"
    description: "高层洞察，少量细节"

Question 3: Number to generate

header: "数量"
question: "生成几张图表？"
options:
  - label: "1 张 (推荐)"
    description: "使用选定风格生成 1 张"
  - label: "2 张"
    description: "生成 2 张，可对比不同风格效果"
  - label: "3 张"
    description: "生成 3 张，多种风格对比"

If the user chooses multiple images: use a different style for each image (recommended style + rotation of other styles), and save them to the same directory.

Step 3: Generate the prompt

Based on the chart type, user content, and selected style, generate a prompt suitable for GPT-image-2.

Prompt structure:

Style prefix - from references/chart-prompts.md select the corresponding one from the style prefix library
Chart Type - Specify the chart type
Content Description - Describe the chart content in detail
Audience Adjustment - Adjust text density and level of expertise based on the audience
Chinese Instructions - Ensure the text in the image is in Chinese
Exclusions - Do not use photography or realistic elements

For prompt templates and style prefixes, see references/chart-prompts.md。

Step 4: Call the API

After confirming the prompt, call the script to generate the image:

python3 /Users/ugreen/.claude/skills/chart-image/scripts/generate_chart.py \
  --prompt "提示词内容" \
  --output "/Users/ugreen/Documents/obsidian/09image/MMDD-图表名.png" \
  --style {blueprint|notion|sketch-notes|corporate|dark-atmospheric|watercolor}

If generating multiple images: call them in order, using a different style for each image.

Step 5: Save and Display

Save path conventions:

/Users/ugreen/Documents/obsidian/09image/MMDD-图表简称/
├── 01-图表名称-blueprint.png
├── 02-图表名称-notion.png
└── ...

MMDD uses the current date
The chart abbreviation is extracted from the title, 2-4 characters
When there are multiple images, append the style name to the filename
After generation, use the open command to display an image preview

Output Specifications

Image Specifications

Aspect ratio: 16:9 landscape (use 21:9 for wide charts such as the Business Model Canvas)
Resolution: 2K
Format: PNG

Content Specifications

The title should be concise and impactful, no more than 15 characters
The text content should be refined, with each point in one sentence and no more than 20 characters
Use specific vocabulary; do not use vague words such as "related" or "various"
Keep numbers and proper nouns accurate

Chart Structure Quick Reference

1 Flowchart

Layout: a chain of nodes from top to bottom or left to right
Core elements: start node, process node, decision node, end node
Connecting lines with arrows, branches with labels

2 Architecture Diagram

Layout: layered (frontend/backend/data/third-party) or circular
Core elements: module blocks, connecting lines, data flow arrows

3 ER Diagram

Layout: entity rectangles scattered, connected by relationship lines
Core elements: entity names, field lists, relationship lines (1:1 / 1:N / N:M)

4 Business Model Canvas

Layout: classic nine-grid (5-column layout)
Nine blocks: Key Partners, Key Activities, Key Resources, Value Propositions, Customer Relationships, Channels, Customer Segments, Cost Structure, Revenue Streams

5 User Journey Map

Layout: horizontal timeline + vertical layers
Core elements: stage names, user actions, emotion curve, pain point annotations

6 Mind Map

Layout: the central theme radiates outward in all directions
Core elements: central node, first-level branches, second-level branches

7 Competitive Analysis Chart

Layout: matrix or table comparison
Core elements: product names, key dimension scores, positioning bubbles

8 SWOT Analysis

Layout: 2x2 matrix
Four blocks: Strengths (S), Weaknesses (W), Opportunities (O), Threats (T)

9 Product Roadmap

Layout: horizontal timeline + vertical module/feature grouping
Core elements: version names, time ranges, feature items, milestone markers

10 Organizational Chart

Layout: tree structure, expanding from top to bottom
Core elements: role names, person names, reporting relationship lines

API Configuration

Configuration item	Value
API URL	`https://api.labnana.com`
API Key	Stored in the `.labnana.env` file
Model	`gpt-image-2`
Provider	`openai`

Common usage

User says	Action
"Draw a flowchart"	Identify type → collect content → choose style → generate image
"Dark-style architecture diagram"	Specify dark-atmospheric style → generate
"Make a SWOT, business style"	Specify corporate style → generate
"Draw a mind map, 3 different styles"	Generate 3 mind maps in different styles
"Regenerate"	Regenerate with the modified prompt