Try it here: https://spatial-captcha-frontend.vercel.app
The idea is simple — humans are naturally good at understanding 3D space, but bots aren't. A bot would need to figure out the spatial orientation of a 3D model rendered in a WebGL viewport, which is a much harder problem than solving text or image classification CAPTCHAs.
Under the hood it uses Three.js for rendering, quaternion comparison on the backend for angle matching (45° tolerance), and the whole thing drops into any site with two lines of HTML. Verification flow is the same as reCAPTCHA — pass token issued on success, server-side /siteverify endpoint.
Curious what you all think — is 3D spatial reasoning actually a durable defense, or will multimodal models eat this for breakfast in a year?