How to build Variational Autoencoder (VAE) using Keras?

Variational autoencoder (VAE) is a generative model, it is used for reconstructing the original images and also generating new images from latent space. Here, the latent space is continuous, so the sampled points will be decoded original images and also produces new images.
in the variational autoencoder (VAE), each image is mapped to a multivariate normal distribution around a point in the latent space.
This multivariate normal distribution is derived from two vectors, they are mu and log_var. These vectors are encoded by the encoder while the input image is being passed through the encoder.
Where, mu - The mean point of the distribution. log_var - The logarithm of the variance of each dimension.

To encode an image into a specific point z in the latent space, the encoder must have to sample this multivariate normal distribution. The decoder will produce similar images from this latent space point z and it’s neighbourhood points.

I am Kishor Kumar Vajja, from aiRobott.