478 points tracked
152 sent · 20 fps
~100 B per frame
Loadit Labs · AI compression
The call that sends meaning, not pixels.
On-device AI reads how your face moves and sends only that. The other phone rebuilds you live from a single photo — about the data of a text chat, instead of a video stream.
Normal video call
≈ 1,200 kbps
Ghost face motion
≈ 13–19 kbps
Video: Zoom’s published 1:1 720p bandwidth. Ghost: our 20 fps test on a talking-head clip. Every call shows its own live numbers.
- One photo, then motionA ~13 KB face photo once. Then only head pose and 152 points, delta-coded.
- Private by designFace data and voice go phone-to-phone, encrypted. Nothing is recorded or stored.
- Honest outputYou see an AI reconstruction, not the camera feed. It is not pixel-perfect.