Xhs Md2img

Convert Markdown text to beautiful Xiaohongshu (XHS) style card images with 5 themes, deterministic browser screenshot rules, auto-pagination, smart title ex...

Convert Markdown text into Xiaohongshu (XHS) style multi-page card images with deterministic browser rendering and screenshot behavior.

Overview

This skill renders Markdown content as publish-ready XHS card images and is designed to run reliably in browser screenshot services.

Use cases:

• Convert long-form content into XHS-ready multi-image posts
• Generate styled card images from Markdown articles
• Produce stable multi-page screenshots with consistent size and pagination

Quick Start

Minimal input — just provide Markdown text:

{
  "markdown": "## 5个提升效率的方法\n\n**1. 番茄工作法** — 25分钟专注 + 5分钟休息\n\n**2. 任务批处理** — 把相似的事情集中做\n\n> 效率不是做更多的事，而是用更少的时间做对的事。"
}

Default export is one 1125x1500 PNG card (375x500 CSS px at export_scale=3).

Input Parameters

See templates/input-schema.json for the full schema.

Core content/style params:

• markdown (required): Markdown content. Use --- for manual page breaks.
• title, author, description: Cover metadata.
• theme: default, monokai, nord, sakura, mint.
• font_family: sans-serif, serif, wenkai.
• padding: small, medium, large.
• show_cover: Whether to show cover block.
• bg_style: ai_art or none.

Browser screenshot control params:

• card_width / card_height: Single card CSS size (default 375x500).
• export_scale: Pixel ratio multiplier (default 3).
• viewport_width / viewport_height: Browser viewport size for rendering stability.
• page_gap: Vertical gap between cards in DOM (default 20).
• pagination_mode: mixed (default), auto, manual_only.
• max_pages: Hard limit to avoid runaway pagination.
• show_page_number: Whether to render 1 / N indicator.
• avoid_orphan_heading: Keep headings with at least one following block.
• last_page_compact: Shrink last card when content is short.

Deterministic Browser Screenshot Contract

Use these hard rules for consistent XHS-like output:

1. Fixed card geometry

• Card ratio must stay 3:4 (375x500 base).
• Output pixel size must be card_width * export_scale by card_height * export_scale.
• Prefer export_scale=3 for XHS quality (1125x1500).

2. Stable browser context

• Run headless browser with deviceScaleFactor=export_scale.
• Keep zoom at 100%; do not use browser print mode.
• Use explicit viewport (recommended 440x760 for default card).

3. Render readiness gates (must pass before pagination/screenshot)

• Wait for document.fonts.ready.
• Wait until document.body[data-fonts-loaded="true"] is present.
• Wait 2 consecutive animation frames with unchanged .xhs-card bounding boxes.

4. Element-only screenshot

• Screenshot each .xhs-card element, never full page + crop.
• Disable animations/transitions during capture.
• Export PNG only.

For implementation runbook and failure checks, read references/browser-screenshot-spec.md.

Rendering Pipeline

1. Smart Format (LLM)

If input is plain text, reformat to Markdown without changing wording:

• Extract a short cover title (10-20 chars)
• Add structure (##, **, lists, quotes)
• Preserve all original symbols/emojis/hashtags verbatim

2. Markdown to HTML

Use python-markdown with extensions:

• tables, fenced_code, codehilite, nl2br, sane_lists, smarty, attr_list, md_in_html, toc
• Convert XHS hashtags (#tag#) into styled pills

3. HTML/CSS Card Construction

Use templates/card-template.html with fixed card dimensions and theme variables.

Card layers:

• .xhs-card: fixed-size card container
• .bg-art: optional low-opacity decorative image
• .card-inner: content layer above background

4. Pagination Rules (Critical)

Apply pagination in DOM, not PDF/print pagination.

Priority order:

1. Split by manual separators (---) first.
2. Within each segment, split by top-level block elements.
3. If avoid_orphan_heading=true, never place a heading as the last visible block on a page.
4. Keep list/table/code blocks intact when possible.
5. If a single block exceeds one page, split safely:

• list: split by li
• table: split by row groups
• code/pre: split by line groups
• paragraph: split by sentences as final fallback

Overflow rule:

• Page content area must satisfy scrollHeight <= clientHeight.
• Keep a bottom safety space (~16 CSS px) to avoid visual clipping.

5. Multi-Page Screenshot Output

For each page:

• Create one .xhs-card node
• Add optional page indicator page / total_pages
• Capture element screenshot in PNG
• Return pages ordered by index

Output shape:

{
  "__type": "xhs_card_images",
  "title": "Extracted or provided title",
  "theme": "default",
  "total_pages": 3,
  "pages": [
    {
      "index": 0,
      "page": 1,
      "total_pages": 3,
      "width": 1125,
      "height": 1500,
      "size_bytes": 123456,
      "url": "https://...",
      "oss_uploaded": true
    }
  ]
}

If OSS is not configured, return data_uri for each page.

XHS Visual Quality Checklist

Before returning images, verify:

• Exact size per page matches export formula
• No cropped text at bottom edge
• Heading/body spacing is consistent across pages
• Page number style is subtle (low opacity)
• Last page has no large empty area when last_page_compact=true

Theme System

5 built-in themes are defined in references/themes.md.

Theme	Background	Feel
default	White #ffffff	Clean, professional
monokai	Dark #272822	Tech, developer-oriented
nord	Deep blue #2e3440	Nordic minimalist
sakura	Soft pink #fff5f5	Warm, feminine
mint	Light green #f0faf4	Fresh, natural

AI Background Generation

When bg_style: "ai_art":

1. LLM generates an abstract decorative prompt from content/theme
2. Provider auto-select:

• Gemini first when LLM_BASE_URL is Google
• DashScope wanx fallback when available

3. Blend into card with low opacity (8-18% by theme)

Constraints:

• Decorative only (watercolor/bokeh/geometric/plant silhouettes)
• No text, no faces, no explicit objects

See references/api-reference.md for provider API details.

Privacy & External Endpoints

Network calls may include:

• LLM API: format + title + bg prompt generation
• DashScope/Gemini: background image generation (prompt only)
• Alibaba Cloud OSS: optional output image upload

No raw user document is sent to image models; only abstract prompt text is used.