I tested a human-only ‘Ghost Font’ that tricks AI readers

Artificial intelligence has grown remarkably adept at scanning documents, deciphering messy handwriting, and even parsing blurry screenshots. Yet a new creation called Ghost Font has surfaced to expose a blind spot most of us never knew existed. This experimental typography project functions as a visual illusion, embedding hidden letters within a sea of moving dots and relying on the human brain’s knack for detecting motion rather than visible shapes. Naturally, I had to dig into how it works and why it seems to stump today’s leading AI models.

Ghost Font is the brainchild of designer Eric Lu, and it explores the fundamental gap between human and machine vision. Instead of drawing letters with solid lines, the project fills the screen with thousands of tiny dots. The dots that form the hidden message move in one direction while the surrounding dots drift in another. Your brain instinctively groups those moving pixels together, allowing the concealed word to emerge even though no visible outlines exist. Pause the animation, though, and the illusion vanishes, leaving only what appears to be random visual noise. The system can even include decoy text designed to mislead AI systems, causing some models to confidently identify the wrong message while humans continue to see the intended one.

Curious whether your favorite AI chatbot can crack it? You can run the same experiment in just a few minutes. Start by visiting the Ghost Font website. Enter a short phrase such as “HELLO HUMAN” or “TOM’S GUIDE.” Generate the animation and make sure you can read the hidden message yourself. Then download the animation or record your screen. Upload it to ChatGPT, Claude, or Gemini and simply ask, “What does this animation say?” Compare the responses. Does the AI identify the hidden message, read the decoy text, or confidently guess something else? For an extra challenge, try again with a more specific prompt like, “Analyze the motion in this animation over time to determine the hidden text,” and see whether giving the AI more context changes the result.

Ghost Font exploits a fundamental difference in how humans and today’s AI vision models process visual information. Humans excel at recognizing motion patterns because our visual system naturally combines tiny changes over time into meaningful objects, letting us spot the hidden letters almost instantly. Many multimodal AI models, however, still analyze video as a rapid sequence of individual images. Because Ghost Font deliberately removes the high-contrast edges and stable letter shapes that optical character recognition systems rely on, the hidden message becomes surprisingly difficult to detect. Let me be clear: this doesn’t mean AI can never read Ghost Font. It simply means the illusion targets a weakness in how many current vision systems handle moving images. So before you start hiding passwords inside Ghost Font animations, there’s an important caveat. Ghost Font isn’t encryption, and its creator doesn’t present it as a security tool. Given enough video frames, optical-flow analysis, or specialized computer vision techniques, AI systems can likely recover the hidden message.

In fact, some developers have already reported success after explicitly explaining how the illusion works or allowing models to analyze the animation frame by frame. As vision models continue to improve, they’ll almost certainly become better at decoding tricks like this.

Ghost Font matters because it highlights something we don’t often consider: humans and AI don’t actually “see” the world the same way. For decades, CAPTCHAs exploited the gap between human and machine perception by generating distorted text that people could still recognize but computers struggled to read. Ghost Font explores a modern version of that idea. Instead of distorting letters, it relies on motion itself as the signal, taking advantage of a capability humans perform almost effortlessly. Ghost Font probably won’t stay ahead of AI forever, but the project offers a fascinating glimpse into the differences between human and machine perception at a time when multimodal AI is becoming increasingly capable. Rather than proving AI is easily fooled, Ghost Font shows that there are still corners of human vision that today’s models haven’t fully mastered.