Sign Chat is an experimental ASL camera plugin. It sends a short sequence of images to ChatGPT, which attempts to interpret your signing and reply. It does not contain a trained ASL recognition model.
https://sign.zentimental.org/mcpEnable the switch to send one phrase on a pause or after 5.5 seconds of movement. Capture then pauses until you read the reply and press Reply read — start next phrase. This is phrase-by-phrase input. Background movement can trigger capture. The app cannot tell when ChatGPT finishes answering.
ChatGPT chooses the model. Pro took 79 seconds for a short sample in our initial test. Choose a faster model setting for testing shorter waits, and check its interpretation carefully. The plugin cannot switch models for you or promise a response time.
The ChatGPT host must grant camera access to the plugin, in addition to your browser and OS permissions. If it does not, open the camera preview, capture a phrase, choose Save phrase image, and attach the downloaded image to ChatGPT manually. Ask: “Attempt to interpret this ordered ASL sequence. Tell me if it is unclear; do not guess.” The preview does not translate by itself. Sampled images can miss important movement.
See the supplied source package and validation notes for automated tests of capture, transport, and failure behavior. Those tests do not establish ASL comprehension. A small photo test found clear, tentative, and partial readings. Live-camera recognition with fluent signers still needs validation. Read the current results and limitations.
Support: chaollapark@gmail.com. Describe the problem and browser version; do not email camera recordings or private chat content.