increasing KL divergence on variational inference with multiple latent variables

Question

I am using variational inference and have three hidden variables. I optimize the ELBO (i.e. likelihood of data, KL divergences corresponding to all 3 latent variables) but o