three.ws avatars fingerspell any word or number in ASL, sign every chat reply, and read your own signing back from a webcam. The whole conversation, both directions, in 3D that runs in the browser.
All 26 ASL letter handshapes plus the numbers 0-9, with the traced J and Z and the double-letter bounce. Type a word in the Animation Studio and export the signing as an animated GLB, or share it as a link.
Open the Animation Studio →
Turn on sign mode in the agent chat and the avatar signs every answer:
library signs where they exist, fingerspelling for the rest. Any embedded
avatar gets it with one sign-language attribute.
Fingerspell into your webcam and the AI reads it to text: landmarks are extracted on your device, so the video never leaves your machine. The demo below runs the same recognizer.
Try the webcam demo ↓A signer spells names and loanwords and signs everything else. These words have real signs in the avatar’s vocabulary: a handshape, a place on the body, and a movement, on one or both hands. Pick one to watch it. Everything else in a sentence still fingerspells.
Click start, fingerspell a word at your camera, then stop. Your browser turns the video into hand and body landmarks locally; only those coordinates are sent for recognition. Webcam fingerspelling is imperfect (expect the odd wrong letter), so review the text before you trust it.
Best results: good light, hand in frame, one letter at a time. Numbers and the full A-Z alphabet are supported.
Honest scope: words with a sign are signed, everything else is fingerspelled letter by letter. Signs are strung together in English word order, which is signed English, not ASL grammar: real ASL reorders sentences, inflects verbs through space, and carries grammar on the face. The vocabulary above is a starting set and is meant to grow. Read the full breakdown.