Type any word and watch it fingerspelled, spliced live in your browser from 26 motion-capture letter takes.
Each letter comes from its own motion-capture take. Rather than cross-fading between static handshapes — which is what makes concatenated signing look robotic — the arm is lifted into signing space once using real captured motion, then only the handshape changes between letters, following a transition profile fitted from all 26 takes.
Articulation is coarticulation: at speed a handshape never fully forms, it is pulled toward its neighbours. That single parameter is most of the difference between fluent and dictionary-crisp.