text-preserve

What This Skill Does

The text-preserve skill is a powerful solution designed to bridge the gap between AI image generation and precise typography for non-Latin scripts, such as Korean (Hangul), Japanese, and Chinese. Standard AI diffusion models often struggle with complex glyphs, resulting in garbled text or nonsensical characters. This skill solves the problem by acting as a pre-rendering middleware: it detects non-Latin scripts, uses Gemini LLM to analyze the prompt for design intent, renders high-quality text onto a transparent PNG using server-side Canvas, and injects that PNG as a precise reference into the final image generation workflow. This ensures that your brand names, posters, and document titles remain legible and professional regardless of the language used.

Installation

To integrate text-preserve into your environment, follow these steps:

1. Ensure you are in your project directory.
2. Execute the setup script: cd {baseDir} && node setup.mjs.
3. Verify that your environment variables are configured, specifically ensuring GEMINI_API_KEY is present.
4. Ensure Noto Sans font files are placed in {baseDir}/fonts/ to allow the Canvas engine to render characters correctly.

Use Cases

This skill is ideal for professionals creating marketing materials where precise wording is required. Common use cases include generating localized promotional posters for festivals, creating social media thumbnails with readable Korean text, designing multilingual informational flyers, and generating visual content for educational tools that require high-fidelity text-to-image integration.

Example Prompts

1. "가을 맞이 서촌 지역 축제 포스터를 만들어줘. 타이틀은 '2024 서촌의 밤'으로 해줘."
2. "'오늘의 메뉴: 떡볶이와 튀김'이라는 문구가 들어간 깔끔한 분식집 인스타그램 광고 이미지 생성해줘."
3. "'겨울 세일'이라는 텍스트가 강조된 세련된 화장품 브랜드 할인 광고 배너 만들어줄래?"

Tips & Limitations

• Tip: For the best results, provide specific design context in your prompt (e.g., "modern style," "playful fonts"), as the Gemini analyze step uses this to determine the visual hierarchy.
• Tip: You can use the pipeline command to automate the entire flow, but using individual steps (detect -> analyze -> render -> generate) allows for custom adjustments to the rendered text layer before the final image generation.
• Limitation: The skill relies on having appropriate font files installed in the /fonts/ directory. Without them, the rendering fallback may fail or default to unsupported system fonts. Ensure your environment has the necessary font licenses and files.