AI Can Generate an ASL Video. But Does It Actually Know the Sign?
I’ve been building SignBridge , a prototype that takes a word and generates a video of someone signing it in ASL. See the SignBridge demo: https://whysignbridge.framer.website I’ve spent years learning ASL and participating in the Deaf community. That led me to explore whether generative video could make individual signs easier to access. What works so far: The prototype can generate videos for…
I have been working on developing SignBridge, a prototype that takes a word and creates a video of someone signing it in American Sign Language (ASL). You can view the demo at https://whysignbridge.framer.website. Having dedicated years to studying ASL and engaging with the Deaf community, I sought to investigate if generative video could make individual signs more accessible.
So far, the prototype has been successful in generating videos for individual words. I have experimented with prompts that directly name the ASL sign and provided more detailed instructions for handshape and movement. However, there are still challenges to overcome. While a video can appear convincing and correctly capture the handshape, position, or motion, it is not always accurate. The results also vary between different generations.
Currently, I am focusing on single words, as sentences present a much more complex problem. I am reaching out to those who have experience in pose estimation, animation, video generation, or sign language technology. I am eager to learn how they would approach making the hand movements more consistent and verifiable. Ideally, the solutions should be able to be checked by Deaf signers, rather than simply relying on the video looking plausible.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
