# Why is the posterior chosen to be normal in variational autoencoder?

Is there any reason for choosing the posterior $$q(z|x)$$ as normal distribution in variational autoencoder? or is it just for convenience?