Hi everyone! I’m Dastan, an 18-year-old solo AI developer from Kazakhstan.
I wanted to experiment with spatial computing, but I don't own an Apple Vision Pro or any expensive sensors. So, I decided to build a solution that works entirely in the browser using just a standard laptop webcam.
I just launched Solifon Air Canvas. It uses AI to track hand gestures in real-time, allowing you to draw in mid-air and interact with 3D models.
You can test it right now in your browser: www.solifonai.online
(Instruction: Click the "New Feature" button in the top menu and allow webcam access).
Since I'm building this completely on my own, I'd really appreciate your help. If you have any cool ideas for new features, feedback on the tracking, or if anyone wants to collaborate and help me improve the code — let's connect!
Hit me up:
Telegram: @DA_ST_A_N
Top comments (2)
Great work! I’m curious—did you use MediaPipe Hands for the gesture detection, and how did you handle the latency when feeding the webcam frames into the TensorFlow.js model? I’m following you for more AI‑web experiments and would love to swap ideas on optimizing real‑time inference.
Hey Frank! Thanks for the follow! Spot on — I am indeed using MediaPipe Hands for the gesture tracking.
Handling the latency was definitely the biggest headache. To keep the FPS acceptable, I had to rely on the TensorFlow.js WebGL backend and downscale the webcam frame resolution before feeding it into the model. I'm also heavily relying on requestAnimationFrame to ensure it doesn't block the main UI thread, but honestly, there's still a lot of room for optimization.
Would absolutely love to swap ideas on real-time client-side inference! Are you mostly working with TF.js right now, or exploring other frameworks?