Ghost Font: An Anti-AI Motion-Based Visual Communication Experiment

Ghost Font uses motion and decoys to obstruct AI perception

Ghost Font is an experimental method of graphical communication designed to be readable by humans but difficult for AI models to decipher. Unlike traditional fonts, Ghost Font does not exist as a static typeface file; instead, it renders messages as video clips where letters are formed by the motion of dots. Because the letters are only visible through movement, a single static frame contains no readable text, rendering standard screenshot-based OCR (Optical Character Recognition) ineffective.

To further obstruct AI agents, Ghost Font employs a layered defense strategy:

• Motion-Based Rendering: Text is embedded in the movement of dots that blend into the background when paused, preventing static image analysis.
• Decoy Messages: Every generated video includes a decoy message. This is intended to mislead AI agents that successfully analyze motion, leading them to identify the decoy as the actual intended message.
• Noise and Camouflage: The system uses visual noise to further obscure the letter shapes from automated detection.

Comparison with traditional anti-OCR fonts

Ghost Font is a spiritual successor to the ZXX font released in 2013. ZXX used noise, strikethroughs, and false marks to camouflage text from OCR software. While ZXX was effective against the software of its time, modern multimodal AI models can now easily read ZXX-rendered text in a single prompt. Ghost Font attempts to solve this by moving from a static 2D plane to a temporal (video) format, assuming that most current AI models analyze videos by sampling individual frames rather than processing native temporal flow.

Technical critiques and AI decoding capabilities

While the project aims to create a "surveillance-proof" visual medium, technical analysis from the community suggests that the protection is not absolute. Several users reported that advanced AI models and custom scripts can bypass these protections:

• Temporal Analysis: Users reported that GPT-5.6 Sol was able to decode Ghost Font by utilizing temporal analysis, optical flow, and vertical-displacement maps to create a motion map that made the text explicit.
• Algorithmic Cracking: Some users demonstrated that the text could be recovered using simple frame subtraction. By taking two consecutive frames and shifting them to minimize differences, the resulting subtracted image can reveal the letter outlines for OCR processing.
• Human Legibility Issues: Some users noted that the text is difficult for humans to read, comparing the experience to "Magic Eye" 3D pictures and reporting eye strain when viewing on mobile devices.

"I can barely read the actual message, and it's about as 'readable' to me as the Magic Eye 3D pictures. Actually I think I have a headache from looking at it on a mobile screen."

Potential applications and future directions

The creator of Ghost Font suggests that this technology could be integrated into CAPTCHA systems to replace current versions that are easily solved by AI. By requiring the perception of motion to solve a challenge, it would create a higher barrier for automated bots while remaining accessible to humans.

Future development goals for the project include:

• Open-sourcing the generation code to allow for community iteration.
• Expanding text capacity to handle longer strings of text.
• Benchmarking AI progress by using the system as a test for when "video-native" multimodal models (which process motion directly rather than as a series of frames) become mainstream.