mage-xhs-images

name

mage-xhs-images

description

Generates Xiaohongshu (Little Red Book) infographic series with 10 visual styles and 8 layouts through a guided workflow: preferences, analysis, two confirmations, outline variants, and sequential image generation. Use when user asks for "小红书图片", "XHS images", "RedNote infographics", knowledge-card carousels, tutorial slides, save-worthy checklists, or multi-image social graphics for Chinese platforms.

Mage Xiaohongshu Infographic Series Generator

Turn source content into a Xiaohongshu-ready multi-image carousel with:

• content analysis
• audience/hook planning
• 3 outline strategies
• 2 confirmation points
• prompt files per page
• sequential image generation

This skill generates Xiaohongshu infographic carousels through a guided workflow, with stricter guidance for text-heavy technical content and providers that are prone to text corruption.

Usage

# Auto-select style and layout based on content
/mage-xhs-images posts/ai-future/article.md

# Specify style
/mage-xhs-images posts/ai-future/article.md --style notion

# Specify layout
/mage-xhs-images posts/ai-future/article.md --layout flow

# Combine style and layout
/mage-xhs-images posts/ai-future/article.md --style notion --layout list

# Direct content input
/mage-xhs-images
[paste content]

Options

Option	Description
--style <name>	Visual style
--layout <name>	Information layout

Style And Layout

Styles

cute, fresh, warm, bold, minimal, retro, pop, notion, chalkboard, study-notes

Layouts

sparse, balanced, dense, list, comparison, flow, mindmap, quadrant

Recommended Pairings

• Knowledge / SaaS / tech / productivity: notion + dense/list/flow
• Tutorials / learning / teaching: chalkboard + balanced/dense/flow
• Personal sharing / emotional stories: warm + balanced
• Strong warnings / avoid mistakes: bold + list/comparison

Reference Files

Presets and visual building blocks

• references/presets/<style>.md
• references/elements/canvas.md
• references/elements/image-effects.md
• references/elements/typography.md
• references/elements/decorations.md
• references/workflows/analysis-framework.md
• references/workflows/outline-template.md
• references/workflows/prompt-assembly.md

Config docs

• references/config/preferences-schema.md
• references/config/first-time-setup.md
• references/config/watermark-guide.md

Reference Loading Guide

Load only the reference files needed for the current stage:

• During preference setup: read references/config/first-time-setup.md and references/config/preferences-schema.md
• During analysis: read references/workflows/analysis-framework.md
• During outline generation: read references/workflows/outline-template.md
• During prompt assembly: read references/workflows/prompt-assembly.md
• When a style is selected: read only the selected preset file under references/presets/
• When layout details are needed: read references/elements/canvas.md
• When text, decorations, or image treatment need refinement: read the corresponding file under references/elements/

Do not pre-load all presets or all element docs. Keep context small and load only the chosen style/layout docs.

File Structure

Each run creates an output directory:

xhs-images/{topic-slug}/
├── source-{slug}.{ext}
├── analysis.md
├── outline-strategy-a.md
├── outline-strategy-b.md
├── outline-strategy-c.md
├── outline.md
├── prompts/
│   ├── 01-cover-[slug].md
│   ├── 02-content-[slug].md
│   └── ...
├── 01-cover-[slug].png
├── 02-content-[slug].png
└── NN-ending-[slug].png

If the target folder already exists, append a timestamp.

Workflow

Progress Checklist

Mage XHS Progress:
- [ ] Step 0: Check preferences (EXTEND.md) ⛔ BLOCKING
- [ ] Step 1: Analyze content → analysis.md
- [ ] Step 2: Confirmation 1 - Content understanding ⚠️ REQUIRED
- [ ] Step 3: Generate 3 outline + style variants
- [ ] Step 4: Confirmation 2 - Outline + style + elements ⚠️ REQUIRED
- [ ] Step 5: Generate images sequentially
- [ ] Step 6: Completion report

Step 0: Load Preferences (EXTEND.md) ⛔ BLOCKING

Do not ask about content, style, layout, audience, or image count until preferences are loaded or created.

Check paths in this order:

test -f .mage-skills/mage-xhs-images/EXTEND.md && echo "project"
test -f "$HOME/.mage-skills/mage-xhs-images/EXTEND.md" && echo "user"

If not found:

1. Run the setup from references/config/first-time-setup.md
2. Save EXTEND.md
3. Confirm save location
4. Continue to Step 1

Supports:

• watermark
• preferred style
• preferred layout
• language
• custom styles

Step 1: Analyze Content → analysis.md

If the source content is not already a file:

1. save it as source.md
2. back up old source.md if present

Then:

1. classify content type
2. analyze hook strength
3. identify audience
4. estimate save/share/comment potential
5. map visual opportunities
6. design swipe flow
7. detect language
8. recommend image count
9. prepare clarifying questions for Step 2

Use:

• references/workflows/analysis-framework.md

Step 2: Confirmation 1 - Content Understanding ⚠️

Display:

• content type and topic
• key points extracted
• tone detected
• source images count

Use AskUserQuestion for:

1. core selling point (multi-select)
2. target audience
3. style preference: Authentic sharing / Professional review / Aesthetic mood / Auto
4. additional context (optional)

Update analysis.md after the answers.

Step 3: Generate 3 Outline + Style Variants

Always create three different options:

• outline-strategy-a.md: Story-driven
• outline-strategy-b.md: Information-dense
• outline-strategy-c.md: Visual-first

Requirements:

• different page structure
• different recommended style
• explicit style_reason
• page-by-page plan
• page count adapted to the strategy

Reference:

• references/workflows/outline-template.md

Step 4: Confirmation 2 - Outline + Style + Elements ⚠️

Show each strategy with:

• strategy name
• page count
• recommended style
• page summaries

Use AskUserQuestion for:

1. outline strategy
2. visual style
3. visual elements

If the user chooses one strategy:

• copy it to outline.md
• update frontmatter with confirmed style and elements

If the user wants a mix:

• merge selected pages into outline.md

Step 5: Generate Images

For each page:

1. save prompt to prompts/NN-{type}-{slug}.md
2. generate the image
3. report progress

Image Generation Skill Selection

Prefer proactively using mage-images-gen as the default image generation backend when image generation is needed and the user has not requested another backend.

Selection order:

1. If user explicitly requests a backend or provider, follow that choice
2. Otherwise, proactively use mage-images-gen
3. If mage-images-gen cannot work because keys/provider support are unavailable, ask before switching to another image backend such as browser-based Gemini Web

When invoking mage-images-gen:

• for normal text-to-image generation, prefer dashscope first, then seedream
• for reference-image generation, prefer the provider order defined by mage-images-gen
• choose the provider/model based on available keys and the content type
• prefer fewer, larger text blocks for providers known to struggle with Chinese typography
• regenerate only affected pages when text quality fails review

For pages that must keep readable text:

• define exact visible copy per page before generation
• split copy into dedicated text zones such as title, subtitle, section header, or bottom bar
• forbid extra words, random letters, pseudo-handwriting, decorative micro-text, and fake UI labels
• for technical or SaaS pages, prefer clean printed sans-serif text over forcing every label into hand-drawn lettering
• for command snippets, formulas, or English product names, place them in wide code-style cards instead of tiny floating labels

Consistency Rule

Generate image 1 first, without --ref.

If the underlying image provider supports reference images:

• use image 1 as the reference image for all subsequent pages

If the provider supports --sessionId:

• generate one session id and reuse it across the whole series

Watermark Rule

If watermark is enabled, append the watermark guidance from:

• references/config/watermark-guide.md

Provider-Specific Text Guardrails

When the selected provider is known to struggle with Chinese typography or dense knowledge cards, especially DashScope / Qwen-style image models:

1. reduce text density aggressively
2. prefer 1 large title + 2-5 short blocks
3. avoid tiny annotations, stickers, labels, and decorative micro-text
4. explicitly forbid extra text, fake words, random letters, and pseudo handwriting
5. convert complex flowcharts into fewer nodes with bigger labels
6. keep technical terms short and repeated consistently
7. if a page has corrupted text, edit that page's prompt first, then regenerate only that page
8. use an exact text whitelist section so the model knows which strings are allowed to appear
9. give every text block its own region instead of describing the page as one dense paragraph
10. for text-rich pages, prefer qwen-image-2.0-pro when available over speed-first models

For text-heavy technical content:

• prefer notion + flow
• use fewer words per image than the source content suggests
• move detailed explanations into the post caption if needed
• if text is mission-critical, use layout-first prompts with EXACT TEXT REGIONS and NEGATIVE CONSTRAINTS

When the selected provider is Seedream:

1. keep the core prompt concise and concrete; prefer staying within roughly 300 Chinese characters for the main instruction
2. prefer 2K or an explicit pixel size aligned to the target aspect ratio for infographic-style pages
3. use at most 1 reference image for normal carousel consistency unless the user explicitly asks for multi-image fusion
4. if using reference images, keep the reference count low and make the subject/style instruction explicit instead of relying on many inputs
5. for Xiaohongshu carousels, prefer generating pages one by one and do not rely on provider-side group-image generation unless the user explicitly asks for it
6. if prompt optimization is available, prefer the higher-quality standard mode over speed-first settings for text-heavy cards
7. if text fidelity still fails, shorten the copy first before switching providers
8. use subject -> layout -> style -> exact text regions -> negative constraints as the prompt skeleton for text-heavy pages

Text QA Loop

After generation, visually review every page for:

• corrupted Chinese characters
• wrong step order
• invented labels
• broken English product names
• tiny unreadable text

If issues are found:

1. tighten the page prompt
2. reduce text count
3. regenerate only the affected pages

Step 6: Completion Report

Report:

• topic
• strategy
• style
• layout
• output directory
• total images
• generated files

Editing Existing Series

Edit

Always update the page prompt file first, then regenerate the page.

Add

Specify insertion position, create a new prompt, generate the new image, renumber following files, and update outline.md.

Delete

Delete the page file, renumber following files, and update outline.md.

Content Breakdown Principles

1. Cover: hook + highest visual impact → usually sparse
2. Middle pages: one core value per page → balanced, dense, list, comparison, or flow
3. Ending: CTA / summary / interaction → sparse or balanced

Notes

• Two confirmation points are required. Do not skip them.
• Prefer concise text over overly faithful but unreadable text.
• Preserve reproducibility by keeping prompts and outlines updated.
• Keep references, prompts, and outlines self-contained so the workflow remains reproducible and maintainable.