Variational Autoencoder applied on a single class on image

Question

I read and saw a lot of tutorials about VAE and all the concept around distribution and sampling . I read that the new generated data is generally picked from the intersecti