An innovative open-source design project has highlighted a critical gap between human visual perception and machine vision processing pipelines. Developed by Eric Lu, founder of the typography platform Mixfont, the browser-based experimental tool titled “Ghost Font” successfully subverts modern optical character recognition (OCR) and text extraction systems while remaining completely readable to human observers. The software bypasses automated scraping by avoiding traditional font files entirely, instead generating localized dynamic noise videos where text is encoded purely through continuous dot animation.
Initial benchmark testing demonstrated that frontier multi-modal artificial intelligence systems, including OpenAI’s GPT-5.6 Sol Ultra and Claude Fable, failed to accurately decipher the hidden content. Because current AI models primarily parse video files by sampling individual static frames rather than interpreting fluid motion, they register only randomized visual noise. Furthermore, the tool embeds integrated decoy text strings that mislead automated systems into generating confident yet incorrect transcriptions.
While the system is not intended as an absolute cryptographic encryption mechanism, computer science researchers note its significant potential for next-generation CAPTCHA frameworks, data privacy safeguarding, and anti-bot verification. Human eyes easily perceive the text as the background and lettering dots move in opposing directions, but individual frames remain indecipherable.
