Pxink is a free, 100% client-side utility built to optimize AI API spend. AI models like Claude and Gemini bill text tokens by density, but vision tokens strictly by image dimensions. Pxink leverages this quirk by wrapping massive text blocks or codebase logs into highly dense 1928px PNGs. The model's OCR reads the text perfectly, but your total token bill drops by roughly 60%. Requires zero setup, no accounts, and your data never leaves your browser.

Hey hunters! 👋 I built PXINK to solve a problem that was quietly draining my own project's budget: massive context window costs. When sending huge datasets, logs, or codebase history to models like Claude or Gemini, text tokenization adds up incredibly fast. PXINK bypasses this by leveraging a structural billing loophole. Since vision tokens are charged flat fees based on image dimensions rather than character density, rendering text into dense, optimized PNGs forces the model to use its OCR slashing total token costs by roughly 60% on heavy context payloads. Best of all, it's 100% free, requires zero server setup, and runs entirely client-side right in your browser to preserve complete privacy. Quick Tip: This hits a sweet spot for large datasets and heavy logs; short prompts are still cheaper to send as raw text!

Smart move leaning on the width-based vision pricing quirk, and the fact that it all runs locally with zero setup makes it actually usable in a real workflow.

Yes @astambuse84875, definitely agreed. 🙌

Optimization towards efficient token utilization is really crucial given the hefty pricing of the AI companies.

Smart move leaning on the width-based vision pricing quirk, and the fact that it all runs locally with zero setup makes it actually usable in a real workflow.

Yes @astambuse84875, definitely agreed. 🙌

Optimization towards efficient token utilization is really crucial given the hefty pricing of the AI companies.

PXINK

Stop overpaying for Claude tokens: Save by picture prompts

Stop overpaying for Claude tokens: Save by picture prompts