 We have a first part, a recognition model and the recognition model will give us a vector of means and also a standard deviation. Now this will effectively give us here a distribution in z space where we have p of z the prior and q of z phi which tells us where we are in this space and now what we'll do is we will take this distribution here and we will sample multiple possible z's using the r-sample mechanism. Now each of these will effectively be fed through the network through a density network as we discussed it before producing a potential image x prime. Now it's your job, fill in the conf VAE, it's going to be very similar to the conf auto encoder conf auto encoder but now with more distributions.