Developer Eric Lu released a video titled "Ghost Font" that went viral immediately, racking up over 17 million views overnight. The "anti-AI font" experiment uses dynamic noise videos to render text: typing characters on the webpage generates a video where pixels forming letters slide upward uniformly while background noise slides downward uniformly, allowing the human eye to understand it through motion perception. When AI views the video frame by frame, it sees only static snow noise. Eric directly tested Claude Fable and GPT-5.6 Sol Ultra; both failed completely, confidently claiming to have decoded information but reading only decoy texts deliberately embedded by the author. GPT-Sol 5.6 Ultra attempted to decode a message written in Ghost Font, reading the hidden decoy message but failing to capture the actual motion-based message. The entire process runs locally in the browser, with no data sent to any server. Ghost Font leverages the Gestalt "common fate" principle: the human visual cortex completes motion perception separation in 0.1 seconds, but paused video shows pure noise in every frame, and frame-by-frame analysis yields nothing. Current multimodal models break video into individual frames to analyze separately, lacking the ability to "perceive motion".