Text to Elegant Image

Version: 1.2.1
License: MIT
Author: Evan Song · github.com/Songhonglei
Repository: https://github.com/Songhonglei/text-to-elegant-image

Converts Markdown or plain text into beautifully formatted high-definition long images. Works by generating an HTML file and capturing a screenshot with a headless Puppeteer browser.

Resource File Structure

resources/
├── styles/               # 18 styles, one file per style (CSS + HTML example + component variable mapping + decorative elements)
│   ├── 01_cyberpunk.md … 18_vivid.md
│   └── _BASE.md          # Common HTML skeleton + layout conventions + open-source CSS tool library reference
└── components.css        # General visual component library (KPI cards/progress bars/comparison bars/donut/flows/timelines, etc.)

Reading Discipline: Read only the files you need. Once the style is determined, read only the corresponding styles/NN_key.md + styles/_BASE.md files; do not read other style files. Read components.css only when the content includes data/flows/comparisons.

Style Selection

There are 18 styles available:

#	File	Style	Visual Keywords	Best For
1	`01_cyberpunk.md`	Cyberpunk (Default)	Dark/Neon/SVG Hex Grid	AI analysis, tech reports, cool posters
2	`02_minimal.md`	Minimal & Elegant	White background/Serif/Whitespace/Noise rice paper	Literary essays, diaries, humanities
3	`03_apple.md`	Apple-Inspired	Light gray/Frosted glass cards/Colorful windows	Productivity tools, retrospectives, methodologies
4	`04_cowork.md`	Cowork Light Tech	Apple gray-white/Tech blue/Frosted glass Hero	Internal sharing, AI tool intros, product manuals
5	`05_newspaper.md`	Newspaper/Magazine	Ivory/Lead type grain/Black & white print	Historical events, timelines, in-depth narratives
6	`06_bloomberg.md`	Bloomberg Terminal	Pure black CRT texture/Orange glowing text	Data reports, metric comparisons, fintech
7	`07_ink.md`	Chinese Ink Scroll	Rice paper ink wash/Cinnabar red/Brushstroke line	Chinese history, culture & art, Eastern aesthetics
8	`08_steampunk.md`	Steampunk	Dark brown/Copper gears/Rivets/Pressure gauges	Industrial history, technological leaps, mechanical aesthetics
9	`09_xhs.md`	Xiaohongshu Style	White background/Official Red/Rounded cards	Knowledge notes, lifestyle guides,干货 (value) sharing
10	`10_morandi.md`	Morandi (Muted Tones)	Warm gray & cream base/Dusty green/Swatch cards/Restrained whitespace	Brand methodologies, retrospectives, humanities essays
11	`11_glass.md`	Glassmorphism	Colorful gradient background/Light orbs/Frosted cards	SaaS websites, event pages, feature showcases
12	`12_palace.md`	Imperial Palace (Forbidden City)	Dark brown wood base/Cinnabar red/Gold characters/Meander seal	Chinese history, traditional festivals, museum exhibitions
13	`13_fresh.md`	Fresh Natural Green	Light mint background/Forest green/Leaf elements	Health & wellness, lifestyle, science & guide articles
14	`14_earthy.md`	Earthy & Natural	Linen weave/Terracotta orange/Hand-stitched card effect	Handcraft coffee, lifestyle aesthetics, brand stories
15	`15_dreamy.md`	Elegant Purple Dreamy	Lavender base/Purple gradient/Starry accents	Beauty & fashion, emotions, event posters
16	`16_macaron.md`	Macaron Pastel	Cream pink base/Polka dots/Candy-colored numbers	Desserts & maternity, lifestyle sharing, holiday greetings
17	`17_carbon.md`	Dark Minimal	Charcoal dark base/Teal green/Terminal line numbers	Technical docs, developer content, data reports
18	`18_vivid.md`	Vivid Gradient	Purple-pink-orange gradient/Fluid color blocks/Gradient outlined cards	Event promotions, youth brands, music & entertainment

Xiaohongshu Style (Style 9) – Two Modes

Style 9 has two built-in sub-modes, specified by the user; if unspecified, Mode A is the default:

Mode	Trigger Words	Vibe
A — Clean & Formal	"Simple"/"Formal"/"Clean"/"Knowledge"	Thin borders, small border-radius, underlined numbers, high breathing space
B — Rich & Lively	"Lively"/"Rich"/"Cute"/"Life"/"Fun"	Light pink, large border-radius, solid circle numbers, wavy decorative line, Footer with CTA text

After generation, inform the user of the current mode and suggest switching:

"Generated in clean mode. If you'd like a lively version, say 'make a lively one'."

Style Selection Guide

When the user doesn't specify a style, automatically recommend based on content type and inform the user of available options:

Content leans toward → Recommended Style
──────────────────────────────────────
Tech/AI/Data Analysis        → Cyberpunk or Bloomberg Terminal
History/Timeline/Narrative   → Newspaper/Magazine or Chinese Ink Scroll (Chinese history)
Industry/Machinery/Tech Leap → Steampunk
Efficiency/Methodology/Summary → Apple-Inspired or Cowork Light Tech
Literature/Diary/Humanities  → Minimal & Elegant or Chinese Ink Scroll
Corporate Internal Sharing   → Cowork Light Tech
Knowledge Notes/Lifestyle Guides/Value Sharing → Xiaohongshu Style
Brand Methodologies/Retrospectives/High-End → Morandi (Muted Tones)
SaaS/Product Website/Event Pages → Glassmorphism or Vivid Gradient
Chinese History/National Style/Museums → Imperial Palace or Chinese Ink Scroll
Health/Eco/Lifestyle → Fresh Natural Green
Handcraft/Coffee/Life Aesthetics → Earthy & Natural
Beauty/Fashion/Emotions/Romance → Elegant Purple Dreamy
Desserts/Mom & Baby/Cute/种草 (Grow grass / recommendation) → Macaron Pastel
Tech Docs/Developers/Eye Comfort → Dark Minimal
Events/Promos/Trendy/Youth → Vivid Gradient

Example recommendation script:

"Based on the content, I recommend 'Xiaohongshu Style' (strong note-taking feel, good for sharing) or 'Apple-Inspired' (more tool-like). There are 18 styles in total; you can also tell me which one you'd like (e.g., 'Imperial Palace', 'Morandi', 'Glassmorphism')."

Keyword matching when users describe style preferences:

"Techy"/"Cool"/"Dark"/"Neon" → Cyberpunk
"Simple"/"White"/"Clean"/"Artistic" → Minimal & Elegant
"Apple-like"/"Modern"/"Product feel" → Apple-Inspired
"Light tech"/"Internal"/"Blue"/"Apple blue"/"Tech blue" → Cowork Light Tech
"Newspaper"/"Magazine"/"History"/"Retro print" → Newspaper/Magazine
"Data"/"Finance"/"Terminal"/"Bloomberg" → Bloomberg Terminal
"Ink wash"/"Chinese"/"Oriental"/"Scroll"/"Ancient" → Chinese Ink Scroll
"Steam"/"Punk"/"Mechanical"/"Copper"/"Gears"/"Retro industrial" → Steampunk
"Xiaohongshu"/"Notes"/"种草"/"Guides"/"Sharing"/"Red" → Xiaohongshu Style (Official Red #FF2442)
"Morandi"/"Muted tones"/"Low saturation"/"Quiet"/"High-end" → Morandi (Muted Tones)
"Glass"/"Frosted glass"/"Matte"/"Transparent"/"Lucid"/"SaaS" → Glassmorphism
"Imperial Palace"/"Palace"/"Red Wall"/"Gold gilding"/"Cinnabar"/"National style" → Imperial Palace
"Fresh"/"Green"/"Natural"/"Mint"/"Eco"/"Forest" → Fresh Natural Green
"Earthy tones"/"Natural wood"/"Terracotta"/"Linen"/"Handcrafted"/"Warm & down-to-earth" → Earthy & Natural
"Purple"/"Dreamy"/"Lavender"/"Romantic"/"Elegant purple" → Elegant Purple Dreamy
"Macaron"/"Pink"/"Pastel"/"Sweet"/"Cute"/"Girly" → Macaron Pastel
"Dark mode"/"Dark"/"Black minimal"/"Eye comfort"/"Techy"/"Developer" → Dark Minimal
"Gradient"/"Vibrant"/"Trendy"/"Youthful"/"Colorful"/"Dynamic" → Vivid Gradient

Layout Modes (Long Image / Poster)

Default output is a long image (560px wide, vertical multi-section). Switch layout when the user says "poster"/"quote poster"/"cover"/"header image":

Layout	Trigger Words	Specs	Layout Notes
Long Image (Default)	"Long image"/"Big image"/unspecified	Width 560px, height auto	Title area + multiple content sections, full content load
Quote Poster	"Poster"/"Quote"/"Share image"	Width 600px, height 800px fixed (3:4)	One core quote enlarged and centered + large whitespace + small source text, content limited to 30 characters
Cover Card	"Cover"/"Header image"	Width 600px, height 800px fixed (3:4)	Large title + subtitle + 2-3 short bullet points, Xiaohongshu cover rhythm

Poster/Cover mode: Choose any of the 18 styles for its color palette and aesthetic. Change .container to width: 600px; height: 800px; display: flex; flex-direction: column; justify-content: center; for fixed height centered layout; the screenshot command MUST pass --fixed-height 800 to bypass height auto-detection, and pass width as 600, like:

node scripts/export_image.js <html> <output.png> 600 --fixed-height 800

Long image mode (default): Do not pass --fixed-height; the script automatically measures the actual .container height and crops excess bottom margin. Frame Margin: Long image mode defaults to adding a 32px uniform frame around the canvas (left/right by widening the canvas + margin:auto, top/bottom by overriding body padding), equal margins on all sides, no content touching the edges—no manual parameter needed. To customize, use --frame <px> (e.g., --frame 48), or --frame 0 to disable the frame and let content touch the edges. Fixed layouts (--fixed-height) already have the container filling the canvas with internal padding and will automatically ignore --frame. Card Background: The .container in all 18 styles defaults to a card visual (background color + border + border-radius/shadow, each style using its own color scheme and aesthetic, e.g., cyberpunk neon border, ink wash paper-style border, Morandi solid color block). This pairs with the frame margin to create a "card floating on canvas" matted effect—no manual parameter needed; it's on by default. To revert to the old flat full-width style (content blending directly into the background, no card border), pass --flat: node scripts/export_image.js <html> <output.png> 560 --flat. ⚠️ For Glassmorphism (11_glass), the card background is the frosted glass effect itself; --flat would eliminate its core visual—not recommended. Cover cards: Consider adding a soft glow decorative orb (top-right or bottom-right, using radial-gradient + absolute positioning) to balance the whitespace from left-aligned content and avoid a large empty right side.

Execution Flow

Step 1: Parse Content Structure

Read the user's content and extract:

Main title, subtitle
Core quote (to be displayed prominently)
Structured sections (comparisons, lists, timelines, etc.)

Visualization Opportunity Scan (Important): Scan the content for key numbers, percentages, rankings, trends, processes, comparisons, and timelines—these should be converted into visual components (KPI cards, progress bars, comparison bars, donuts, process steps, timelines, see comments in resources/components.css) rather than being presented as plain text. 2-3 key numbers warrant a KPI card; rankings/comparisons warrant comparison bars. Rich visuals are key to a quality long image.

Step 2: Read Style Resources and Generate HTML

After determining the style, read resources/styles/<NN_key>.md (containing style CSS, HTML structure examples, component variable mapping, font injection, signature decorative elements) and resources/styles/_BASE.md (common skeleton + layout conventions)
If Step 1 revealed visualization opportunities, read resources/components.css, copy the required component CSS segments into <style>, and also copy the :root block of the 'component variable mapping' from the style file (components use --t2e-* variables for color matching)
If the style file contains a "Font Injection" section, add the font <link> into <head>
If the style file contains a "Signature Decorative Elements" section, prioritize using these elements to enhance identity
Inject the content into the skeleton, generate the complete HTML, and save it to /tmp/elegant_image_<timestamp>.html using the write tool (the tool name is write, not write_to_file)

Footer Handling Rules (Important):

User has not specified author info: Write the default value into the .footer element in the HTML: <skill name> @ <current model name>
- Example: text-to-elegant-image @ claude-4.6-sonnet
User has specified author info (e.g., "Author: John Doe"): Write the user-specified text.
User explicitly says no footer: Do not generate the .footer element in the HTML; append --no-footer to the screenshot command.

Do not write "Generated by " or any AI identity identifier in the footer.

Step 3: Dependency Check & Initialization (Before Each Use)

Run the setup script to check the environment:

bash <skill directory>/scripts/setup.sh

Checks Node.js (>= v18), Chrome/Chromium, puppeteer-core - three dependencies.
If all are ready, it passes directly and prints ✔ All dependencies ready.
If Node.js / Chrome is missing: The script prints an error and exits (exit 1), prompting to install the missing dependency (Node.js >= v18 / Chrome or Chromium).
If puppeteer-core is missing: The script prompts "Installing..." and runs npm install (the installation output is visible; it only installs the puppeteer-core rendering dependency).

The script returns very quickly when dependencies are ready, with no performance overhead. It is recommended to run it before each use.

Step 4: Emoji Check (Mandatory Before Screenshot)

Absolutely prohibit the use of emoji when generating HTML—the headless environment lacks color emoji fonts, which will render as black square garbled characters. Alternative solutions:

Original emoji usage	Alternative
Section icons (chart/crystal ball/map, etc.)	Plain text title, differentiated by `border-left` color bar
Status markers (checkmark/cross/warning)	Use colored badge: `<span class="badge-green">Fully Automatic</span>`
Decorative symbols	Delete, or use style-specific signature decorative elements (leaves/stars/seals, etc.)
Progress/arrows	Use CSS `::before` pseudo-elements or plain text `->`, `·`

After generation, you MUST run a check:

python3 <skill directory>/scripts/check_emoji.py <HTML file path>

If it FAILs (exit 1) → replace according to the output line numbers → re-run until PASS before proceeding to Screenshot.

Step 5: Execute Screenshot Export

node <skill directory>/scripts/export_image.js \
  <absolute HTML file path> \
  <output directory>/<filename>.png \
  [width] [--author "Author Info" | --no-footer]

Output Directory Configuration (Guide the user on first use):

The output directory is resolved with the following priority (high to low):

The output path explicitly passed in the command
The environment variable T2EI_OUTPUT_DIR
Default value: ./output (the output/ directory under the current working directory)

💡 First-Time User Guide: The first time you generate a long image for a user, ask once: "Which directory would you like to save the images to?"
User doesn't specify → Use the default ./output.
User specifies a fixed directory → Suggest persisting it via an environment variable:
echo 'export T2EI_OUTPUT_DIR="$HOME/my-images"' >> ~/.bashrc
Do not ask again afterward.

Footer Parameter Explanation:

Parameter	Effect
Not passed (default)	Uses the footer already written in the HTML (see Step 2)
`--author "Custom text"`	Replaces the footer with the specified text
`--no-footer`	Completely removes the footer

The script automatically detects the Chrome/Chromium path (Linux/macOS/WSL), automatically waits for webfonts to load (up to 8s; non-blocking if CDN is unreachable), automatically measures content height for precise cropping, and automatically creates the output directory if it does not exist. Width defaults to 600px, with 2x deviceScaleFactor for high-definition output.

Step 6: Return Results and Deliver Image

After a successful screenshot, simply return the local PNG path to the user:

Image generated 🎉
Local path: <output directory>/<filename>.png

The user can view this PNG directly in the file system / editor or distribute it themselves.

📌 Regarding Auto-Distribution (Reserved, Not Built-in) This open-source version only handles the "text → elegant long image" generation; it does not include any automatic image posting capability to IM channels. To automatically push the generated image to channels like WeChat / Telegram / Slack, you can integrate the respective platform's Bot / API after this step (future versions may provide optional adapters). Currently, please let the user manually retrieve the local PNG for distribution.

Style Change (Low-Cost Path)

When a user says "Change to [another] style" for an already generated long image: Do not rewrite the entire HTML. Read the new style file, only replace the <style> segment for the style CSS and the component variable mapping (--t2e-* :root block), and batch-replace the old style's prefix class in the HTML (e.g., mo-*) with the new style's prefix (e.g., fr-*)—the skeleton components for each style are isomorphic (header/title/subtitle/section-title/card/item/num/quote/tag/divider/footer), so they map one-to-one. Then re-run Steps 4-6.

Fallback Mechanism

If the screenshot fails (e.g., browser not installed), the script will output the HTML file path and guide the user to open it in a browser and manually print/save as PDF/image.

Key Notes

The third argument of export_image.js can specify the width (default 600px), with 2x deviceScaleFactor ensuring high-definition output.
Do not reference external images; use pure CSS shapes or inline SVG data URIs for icons to avoid network loading failures. The only allowed external resource is the font CDN (fonts.googleapis.com, a public service; falls back to system fonts, non-blocking).
Output path: <output directory>/<filename>.png (default output directory is ./output, configurable with the T2EI_OUTPUT_DIR environment variable, see Step 5).

Script List

Script	Purpose
`scripts/setup.sh`	Dependency check (Node/Chrome/puppeteer-core); prompts and installs if puppeteer-core is missing (output visible)
`scripts/check_emoji.py`	Emoji check before screenshot; outputs character and line number on FAIL
`scripts/export_image.js`	Puppeteer screenshot export: font waiting + precise height cropping + uniform frame margin + card background toggle (--flat) + footer injection/removal

Songhonglei/text-to-elegant-image

What is text-to-elegant-image?

Ask in your favorite AI

Documentation