If you create text-and-image content on RedNote, WeChat Official Accounts, or other feed platforms, you have most likely hit this problem: the writing is solid, but the images read as AI-generated at a glance — big color blocks, emoji pileups, and the same cookie-cutter Tailwind look.

guizang-social-card-skill is an open-source Skill built specifically for text-and-image creators, with one goal: card images you cannot tell were made by AI. It ships a magazine-grade layout engine, auto-adaptation rules for 11 content categories, free commercial-use image library integration, and a screenshot beautification system.

GitHub: https://github.com/op7418/guizang-social-card-skill

The Problem It Solves

The vast majority of AI text-and-image card tools share three chronic flaws:

1. One-note styling —— whether you are writing film reviews or product reviews, the cards all come out looking the same
2. Text-over-image fails —— white text on light backgrounds, text covering faces, or lines stretching across the whole image and wrecking the composition
3. Awkward image sourcing —— upload your own, drop in an emoji, or generate an illustration that reads as AI at a glance

This Skill solves each of them.

Core Features, Broken Down

Auto-adaptation across 11 content categories

Feed it the same text: call it a film review and you get a movie-poster-style card; call it a product review and you get a comparison shot framed in a device bezel. It ships adaptation rules for 11 common text-and-image categories — travel, career, film and TV, gadgets, books, food, and more.

Each category has its own visual language: warm magazine styling for travel, a dark grid look for career know-how, and high-saturation full-bleed photos for food explorations.

It even built a map component for travel bloggers: pin shop locations and travel routes on it, and the AI generates the annotations automatically.

Text over images, handled in three steps

How text sits on top of an image is the hardest part of a text-and-image card. This Skill does it in three steps:

1. Detect the subjects —— faces, products, text-dense areas; the layout steers around them automatically
2. Compute the landing color and luminance —— deciding text color, whether to add a mask, and how deep the shadow goes
3. Adaptive font size and line breaking —— sized dynamically to the landing area instead of hard-coding a size and letting it overflow

Auto-illustration from three free image libraries

It integrates Pexels, Unsplash, and Wallhaven, three free commercial-use image libraries, by default. Based on the semantics of each body paragraph, the Skill dispatches search terms automatically, fetches images, and crops them to fit the layout, avoiding cuts through faces or main subjects.

The screenshot beautification kit

For app screenshots, chat logs, and product UIs, the Skill automatically adds a macOS / iOS-style device frame and pairs it with backgrounds in different textures (grid paper, dot matrix, warm white, or dark), making screenshots look like official product promo art.

Two visual systems + 28 layout skeletons

• Magazine style: Generous whitespace, big serif headlines, and asymmetric layouts, like a New Yorker cover
• Grid style: Strong grids, sans-serif type, and a geometric feel, in the Swiss graphic design lineage

The 28 layout skeletons are curated from magazines, posters, album covers, and movie posters. The AI only needs to "fill in" a proven skeleton, and the consistency of the finished cards improves dramatically.

How to Install

In your Codex, OpenClaw, Claude Code, or WorkBuddy, just say:

Install this Skill for me: https://github.com/op7418/guizang-social-card-skill

Supported output ratios: 3:4 (the main battleground, optimized for mobile feeds), 21:9, and 1:1.

Who It Is For

• Text-and-image creators on RedNote / WeChat Official Accounts
• Operators who need to make card images frequently
• Content teams unsatisfied with the quality of AI-generated images