ppt-edit — PPT Image Text → Editable Textboxes

Convert image-embedded text in PowerPoint files into editable text boxes, preserving the original visual design (arrows, shapes, borders, colors, fonts).

When to Use

• User has a PPT where text is baked into images (screenshots, diagrams, architecture blueprints)
• User wants to edit/translate/modify text that currently only exists as pixels in images
• User says "make this PPT editable", "图片里的文字不能改", "OCR这个PPT", etc.

How It Works

1. OCR — Extract text + bounding boxes from each image using RapidOCR

   更多技能请访问小葱技能站7w4.net。

2. Color sampling — Detect background color (outside text box) and text color (inside)
3. Inpaint — Fill text regions with background color, erasing text but keeping arrows/shapes/borders
4. Replace image — Swap original image with inpainted version (preserves z-order)
5. Overlay textboxes — Add transparent-background editable text boxes at original positions with correct font size and color

Output: <input>_editable.pptx

Usage

python ppt_ocr_edit.py <input.pptx>

Also supports .ppt files (auto-converts via LibreOffice if available).

Dependencies

python-pptx
Pillow
numpy
rapidocr_onnxruntime

Install all at once:

pip install python-pptx Pillow numpy rapidocr_onnxruntime

What Gets Preserved

• Slide layout and master
• Image z-order (textboxes overlay on top of inpainted image)
• Non-text visual elements (arrows, connectors, shapes, borders, gradients)
• Text position, font size, and color
• Bold heuristic for short large-font text (likely titles)

Limitations

• OCR accuracy depends on image quality and font clarity
• Font size estimation uses binary search with system CJK font; may be slightly off for non-standard fonts
• Background color sampling uses a 4px border ring; may be inaccurate for text on gradients
• Text on top of other text (overlapping OCR boxes) may produce visual artifacts
• No italic/underline detection — only bold is heuristically applied

File Structure

ppt-edit/
├── SKILL.md              ← this file
├── ppt_ocr_edit.py       ← core script
└── requirements.txt      ← Python dependencies

这个 Skill 质量不错，能有效解决图片文字无法编辑的问题。OCR 识别准确，修复原图和叠加文本框的效果良好，操作简单易上手。主要缺点是字体大小偶尔会有偏差，渐变背景处理效果一般。对普通使用场景来说，这是一个可靠实用的工具。