The text you see is composed of noise pixels that slide in a consistent direction inside the letter shapes. Each individual frame is pure noise, but the directional motion creates a pattern you can see.

Your visual system integrates this motion across approximately 200 milliseconds. Your brain tracks the directional flow of pixels across multiple frames, and this accumulated motion resolves into readable letterforms.

AI vision systems and screenshots analyze single frames at a time. Since each frame contains purely noise with no spatial structure, there's nothing to detect. The text exists only in the temporal dimension.

The effect uses a feedback buffer system where each frame reads data from the previous frame. Inside text regions defined by a mask, pixels copy their color from a neighboring pixel in the slide direction. At the boundaries, fresh random noise is injected.

This creates a continuous flow of noise that moves coherently inside the text shapes. Your brain's ability to track motion over time makes this directional flow appear as solid letterforms.

Higher difficulty levels vary the effect, making the motion harder to track and requiring longer temporal integration to read.

This technique reveals a fundamental perceptual asymmetry: current AI vision excels at spatial analysis but doesn't naturally integrate temporal information the way biological vision does.

As AI vision becomes ubiquitous, exploiting these differences could enable new forms of privacy and human-only communication. Information could exist in formats that are invisible to screenshots, recordings, and AI analysis, but perfectly clear to human observers.