SEEDOC - Quest for the Golden Ratio

🎯 The Day We Found the Golden Ratio

Remember why we chose Vision JPG? To discover the Ultra Compact Golden Ratio (1600px, Q60, 65000px, line_spacing +1px), we needed countless experiments.

That's when we created SEEDOC - a simple app that converts MD files to JPG/PNG/WEBP/TIFF and measures how accurately Vision AI reads them.

Original Purpose: Testing DeepSeek OCR Actual Result: Proof that JPG format is optimal for Vision 💪

💡 Why We Needed SEEDOC

Problem: Difficulty Measuring Quality

Vision AI optimization must satisfy all of these simultaneously:

• Readability: AI accurately recognizes text
• File Size: Token efficiency (smaller is better)
• Speed: Conversion time (faster is better)

Testing hundreds of setting combinations required a repeatable tool.

Solution: SEEDOC's Core Features

✅ 4 Format Support: JPG, PNG, WEBP, TIFF ✅ Quality/DPI Control: 50-100 quality, 72-300 DPI ✅ Batch Processing: Multiple file conversion ✅ OCR Reverse: Image → MD (accuracy verification)

🔬 Experiment Process

Step 1: Format Comparison

Test Document: 5000 lines of code
Conversion Settings: 1600px, Q60~100, DPI 150~300

Results:
- JPG: 232KB, Vision readability 100%
- PNG: 1.2MB, Vision readability 100%
- WEBP: 180KB, Vision readability 95% (slight blur)
- TIFF: 3.5MB, Vision readability 100%

Conclusion: JPG wins in size-to-quality ratio 👑

Step 2: Discovering the Golden Ratio

After hundreds of tests with SEEDOC:

• 1600px width: Minimum width maintaining readability
• Q60: Sweet spot between compression vs quality
• 65000px height: Optimal size for Vision single page
• line_spacing +1px: 19% savings in line spacing

232KB → 3 pages = 5,400 tokens (88% savings!) 🚀

🛠️ How to Use SEEDOC

Basic Usage

1. Add Files: Drag & drop (MD, TXT, code files)
2. Adjust Settings:
   • Format: Select JPG
   • Quality: 60
   • DPI: 200
3. Auto Convert: Automatic processing when added to queue
4. Check Results: Generated in same folder as original

OCR Reverse (For Verification)

Drag an image to generate MD file via Tesseract OCR:

• Auto-detect Korean/English
• Preserve code blocks
• Maintain table structure

Purpose: Verify Vision AI reading accuracy

📊 Performance Comparison

<div className="grid grid-cols-5 gap-3 my-0 text-sm"> <div className="font-bold border-b-2 border-primary pb-2">Format</div> <div className="font-bold border-b-2 border-primary pb-2">Size</div> <div className="font-bold border-b-2 border-primary pb-2">Vision Readability</div> <div className="font-bold border-b-2 border-primary pb-2">Speed</div> <div className="font-bold border-b-2 border-primary pb-2">Use Case</div> <div className="font-bold">JPG</div> <div>⭐⭐⭐</div> <div>⭐⭐⭐⭐⭐</div> <div>⭐⭐⭐⭐</div> <div className="text-primary font-semibold">Vision AI (Optimal!)</div> <div>PNG</div> <div>⭐</div> <div>⭐⭐⭐⭐⭐</div> <div>⭐⭐⭐</div> <div>Archiving</div> <div>WEBP</div> <div>⭐⭐⭐⭐</div> <div>⭐⭐⭐⭐</div> <div>⭐⭐⭐⭐</div> <div>Web Sharing</div> <div>TIFF</div> <div>⭐</div> <div>⭐⭐⭐⭐⭐</div> <div>⭐⭐</div> <div>Professional</div> </div>

🎬 Conclusion

SEEDOC isn't a flashy app. But it's the unsung hero that accomplished the critical mission of discovering the golden ratio.

Thanks to it, we achieved:

• Established optimal Vision JPG settings
• 90% token savings
• Large codebase analysis capability

A simple tool, but with enormous impact. 💕

🔗 Links

• Download: SEEDOC v1.0.0 (78MB, Windows Portable) - No installation needed, run immediately!
• GitHub: <a href="https://github.com/tau-breath/SEEDOC" target="_blank" rel="noopener noreferrer">tau-breath/SEEDOC</a>
• Tech Stack: Electron + Sharp + Tesseract.js

P.S. We now use different scripts for MD-JPG conversion, but without SEEDOC, there would be no golden ratio. Thank you, SEEDOC! 👋