A proof-of-concept captcha that relies on movement rather than static distortion or objects recognition to distinguish humans from AI vision tools.
Static captchas (twisted text, image grids, reCAPTCHA challenges) are increasingly solved by vision models (OCR, GPT-4V, Claude, etc.). This prototype explores a different angle: temporal reasoning. Users must track which characters move along a geometric path over time, while decoy characters move differently. This combination of motion and spatial awareness is harder/more expensive for frame-by-frame AI analysis.
- Solution letters and noise characters are scattered in a rectangle.
- Solution characters move along a shared shape (circle, square, triangle, etc.).
- Noise characters move according to a separate pattern (random jitter, square, circle, etc.).
- The user observes, identifies the solution characters by their movement signature, and types them in any order.
A static frame taken independantly would be useless:
But as soon as you get some movement, the solution becomes visible:
Open index.html in a browser. No server or build step required.
- Client-side only – Solution is visible in memory; a proper implementation would require server-side generation and validation.
- No accessibility review – Motion-based challenges are inherently difficult for users with vestibular disorders, seizure susceptibility, or screen readers. A live captcha would need alternatives (e.g. reduced motion, audio fallback).
- Not peer reviewed – Security and usability have not been evaluated by third parties.
This prototype was created with the assistance of generative AI (Composer 1.5).