Real Time model Deployment

i made a simple flask server to deploy my realtime face detection model using http requests
but it was very slow 3 or 2 fps for 400x300x3 image
so i’m asking what is best way to communicate with server (to be fast as video chat)