Multiplatform Post Illustrations

Create a traceable illustration set from written content.

When to use it

Illustrate an article

I have finished content for WeChat, Zhihu, Weibo, or Toutiao. I want the strongest ideas selected and illustrated one image at a time.

Create a carousel

I have a Xiaohongshu note and need a coordinated cover and content carousel rather than unrelated decorative images.

Continue an image set

I need to add, restore, rebrand, or regenerate one image in an existing bundle without rebuilding the accepted images.

Won't take

This is not for photo retouching, product mockups, photoreal campaigns, exact long text inside images, or another explicitly named image skill.

What it produces

Style is chosen once for the set, while each image carries only one core idea.

• Planning files: Saves content analysis, a mandatory shot list, and one prompt per image
• Image assets: Generates and checks each raster separately, preserving accepted native pixels
• Branding: Applies the real brand asset deterministically only when the resolved policy enables it
• Bundle record: Writes manifest.md with backend, geometry, attempts, file paths, QA, and remaining risk
• File changes: Creates or updates post-illustration-output/<content-slug>/ in the user's project
• External activity: Uses one verified image backend and may incur generation cost
• Never: Lets the model draw logos, stretches accepted output, or writes generated assets into the skill directory

Prerequisites & boundaries

Prerequisites

Provide readable source content and a target platform: WeChat, Xiaohongshu, Zhihu, Weibo, or Toutiao. A registered platform style and a verified image backend must be available.

Neighbor skill split

Need	Hand off to
Build a reusable style	visual-builder
Make one WeChat cover	wechat-sketch-cover
Run the full pipeline	content-production

Scenarios it declines

• Natural photography, portraits, product retouching, or exact campaign recreation
• Images that require long, perfectly rendered copy
• An unknown platform style without user confirmation of a registered fallback

Subtle edges

• A requested count is a ceiling, not permission to invent filler images
• Branding follows user override first, then the selected style default
• A failed ratio is retried at the verified request size; accepted images are never resized to force compliance