Ski Face Swap — Neural v2 (deepfake-grade)
Real neural swap (inswapper) running on your device's GPU. Watch FPS + latency.
-- fps
latency-- ms
detect-- ms
swap-- ms
starting…
Loading neural models…
0%
First load only (~430MB, then cached). Keep this tab open.
Loading models…
Neural = real deepfake swap (inswapper) — higher fidelity, heavier. Fast = the light landmark-warp from v1, for comparison. If Neural's FPS is too low on your phone, the gap between the two tells us exactly how much headroom the neural model needs.
Target on your phone: 15+ fps for Neural is great (deepfake models are ~50× heavier than the light warp). Even 8–12 fps can look smooth on a call with frame-smoothing.
Target on your phone: 15+ fps for Neural is great (deepfake models are ~50× heavier than the light warp). Even 8–12 fps can look smooth on a call with frame-smoothing.