Xiaohongshu Post Imagegen

Xiaohongshu Post Imagegen

Overview

Turn a short topic into a ready-to-post Xiaohongshu carousel: multiple vertical images with Chinese copy rendered inside the images, title options, one unified whole-post caption that matches the image set, per-slide notes, and tags.

If the primary input is a WeChat screenshot folder, stop and route the task to wechat-chat-to-xiaohongshu. Keep this skill for topic-first, fully generated carousels.

This skill is normally used together with the built-in imagegen skill/tool. If the task involves generating raster images, follow the imagegen skill rules and use built-in image_gen by default.

Default Output

Unless the user specifies otherwise, create 4 portrait images:

1. Cover: brand/topic hook and core offer.
2. Content: what the reader can learn or gain.
3. Audience: who it is suitable for.
4. CTA: next step, private-message keyword, or action prompt.

Use a Xiaohongshu-friendly vertical layout. Prefer 3:4 portrait output. Keep in-image Chinese text short, large, and easy to read on mobile.

Always include one unified publishing copy package for the entire carousel, not only per-image notes.

Workflow

1. Extract the topic, offer, audience, brand name, and CTA from the user request.
2. If the user only provides a topic, infer a practical 4-slide structure. Do not ask follow-up questions unless a required brand or compliance constraint is missing.
3. Draft exact in-image Chinese copy for each slide before image generation.
4. Generate each distinct slide with one separate built-in image_gen call. Do not use one generic prompt for all slides.
5. Inspect every generated image with view_image when possible. Check:
   • Chinese characters are correct and simplified.
   • No text is cropped, tiny, or unreadable.
   • No unwanted QR code, phone number, watermark, lorem ipsum, or random English appears.
   • The slide role matches the requested carousel structure.
6. If a slide has a visible text error, regenerate only that slide with a narrower prompt and a stronger verbatim-text constraint.
7. Save final images into the current workspace under 成品/<safe-topic-name>/. Keep generated originals in place.
8. Create 小红书发布文案.md in the same output folder with:
   • Image filenames.
   • In-image copy for each slide.
   • Per-slide caption notes.
   • Title options for the post.
   • One unified whole-post caption written for the full image set.
   • Image order and how each image supports the caption.
   • Comment/DM call to action.
   • Tags grouped by core topic, audience, scenario, and long-tail search.
9. Report final saved paths, the generation mode, and any caveats about image text quality.

Unified Caption Rules

Read references/copywriting.md before writing the publishing copy. Apply those patterns without promising virality or fabricating outcomes.

The unified caption should:

• Start with a concrete pain point, result, or question that matches the cover.
• Explain why the carousel is worth saving or reading.
• Reference the image order naturally, such as "这 4 张图整理了...".
• Use short paragraphs and scan-friendly bullets.
• Include a soft CTA such as "想要路线，私信：AI学习".
• Avoid exaggerated claims such as guaranteed income, guaranteed admission, guaranteed job offers, or impossible learning timelines.

Prompting Rules

Read references/prompt-templates.md for slide prompt templates when generating a carousel.

Use these constraints in every image prompt:

• Use case: ads-marketing.
• Asset type: Xiaohongshu carousel post illustration.
• Text must be rendered verbatim in Simplified Chinese.
• Include only the listed text; avoid extra text.
• Use clean bold Chinese sans-serif typography.
• Use a balanced palette such as warm white, coral red, teal green, charcoal black, and yellow accents.
• Avoid dominant purple/blue gradients, QR codes, phone numbers, watermarks, lorem ipsum, and random UI text.
• Keep text inside safe margins and readable on mobile.

For Chinese words that often drift, explicitly specify the exact character. Example: 对 must be simplified Chinese 对 (U+5BF9), not 対.

File Naming

Use stable filenames:

• 01-封面-长期招生.png
• 02-课程内容.png
• 03-适合人群.png
• 04-私信转化.png

Adapt the suffixes when the topic is not enrollment-related, but keep numeric prefixes.

If replacing a flawed generated image, move the flawed copy to _备份/ instead of deleting it.