Why is live 2.4 seconds behind me?
Updated 3 Nov 2026 · 4 min read
The live model draws 12 frames at a time. Before it can start a chunk, it waits for those 12 frames from your camera; then it generates them on the GPU and sends them back to you.
Where the time goes
Filling a 12-frame chunk takes about 0.45 s, generating it about 1 s, and the network and playout most of the rest. Together that’s about 2.4 seconds.
Tip: pause briefly after a gesture. Viewers see it a moment later, so a short pause keeps your words and movement together.
Keeping in sync
Audio isn’t processed, so your voice arrives before your lips. In OBS, delay your microphone by 2.4 s in the audio settings.
Will it get faster?
Yes, over time: shorter chunks and lighter models bring the delay down. We’ll update this page when they ship.
Was this helpful?