logo
Kurashizu Blogwhere ideas flow
  • Home
  • News
  • Blog
  • About
  • Home
  • News
  • Blog
  • About

© 2026 Kurashizu. All rights reserved.

Admin·New Post·Service Status
Powered by Cloudflare·01.11.26
← Back to news

Ghost Font: A font that humans can read but AI cannot

Original article
mixfont.com|142 points|108 comments|by justswim|Jul 11, 2026

Ghost Font: A Visual Cipher for Humans, Not AI

Mixfont

Ghost Font is an innovative "anti-AI" communication tool designed to be legible to the human eye while remaining indecipherable to the world's most advanced artificial intelligence models.

👁️ What Exactly is Ghost Font?

Unlike a standard .ttf or .otf file, Ghost Font isn't a traditional typeface. Instead, it is a graphical experiment in communication that utilizes motion to convey text. By blending video, noise, and strategic decoys, it creates a medium where humans can easily spot a message, but AI agents struggle to perceive it.

How the Magic Works

The system doesn't rely on static shapes. Instead, it uses a series of moving dots.

  • Human Perception: Our brains are wired to detect patterns in motion, allowing us to read the letters as they shift.
  • AI Perception: Most AI models analyze videos as a sequence of static frames. Because the dots are designed to blend into the background in any single frame, the message vanishes.

Key Insight: Because the letters only exist between frames, taking a screenshot of the page is useless. A static image reveals nothing.


🤖 Testing Against the Giants

The creator tested Ghost Font against cutting-edge models, including Claude Fable and GPT Sol 5.6 Ultra. Even these agents, which possess advanced coding capabilities, failed to decode the messages unless they were explicitly told the exact technique used to hide the text.

The Results

In most cases, the AI fell for the "trap."

GPT-Sol 5.6 Ultra failure GPT-Sol 5.6 Ultra identifies the decoy but misses the actual moving text.

Claude Fable failure Claude Fable (Max Reasoning) similarly falls for the decoy message.

When ChatGPT 5.5 Pro attempted to analyze a Ghost Font clip, it spent 19 minutes processing the data, only to eventually hallucinate a message that wasn't there.


📜 Evolution: From ZXX to Ghost Font

This project is a spiritual successor to the ZXX font released by Sang Mun in 2013. ZXX attempted to defeat Optical Character Recognition (OCR) using camouflage and noise.

ZXX Font Example

While ZXX was surveillance-proof effective in 2013, modern AI has evolved. Today's multimodal models can slice through ZXX's noise with ease.

Comparison Table: ZXX vs. Ghost Font

FeatureZXX Font (2013)Ghost Font (Current)
MechanismStatic noise & camouflageTemporal motion & decoys
FormatTraditional TypefaceVideo/Motion Graphics
AI VulnerabilityEasily read by ChatGPT 5.5Highly resistant to frame-analysis
Screenshot ProofNoYes

ChatGPT reading ZXX Example: ChatGPT 5.5 effortlessly reads ZXX text in a single prompt.


🛡️ The Layered Defense Strategy

Simply hiding text in motion isn't a perfect shield. A sophisticated agent with a local code execution environment could potentially analyze the pixel shifts over time to reconstruct the letters.

To counter this, Ghost Font employs a Decoy Layer.

  1. The Primary Message: Hidden in the motion of dots.
  2. The Decoy Message: A visible but "fake" hidden message.

When an AI searches for a secret, it finds the decoy first and assumes its search is complete, effectively ignoring the actual moving message.

The Logic of Detection

We can represent the probability of AI detection P(D)P(D)P(D) as a function of the decoy's effectiveness EEE and the model's reasoning RRR: P(D)≈RE×motion_complexityP(D) \approx \frac{R}{E \times \text{motion\_complexity}}P(D)≈E×motion_complexityR​


🚀 Future Implications & Goals

While true security always requires encryption (a key known only to humans), Ghost Font explores the boundaries of visual perception.

Potential Applications

  • Next-Gen CAPTCHAs: Replacing static distorted text with motion-based challenges that bots cannot solve.
  • AI Benchmarking: Using these videos to test when "video-native" models (rather than frame-based models) finally emerge.

Roadmap for Development

  • Open-source the video generation code.
  • Increase the maximum character limit for longer strings.
  • Improve human legibility (as it is currently quite difficult to read!).

ChatGPT 5.5 Pro Hallucination

Ultimately, as AI begins to dominate font generation and visual synthesis, projects like Ghost Font serve as a reminder of the unique way humans perceive the world—and a way to preserve a uniquely human creative voice.