# In variational inference, is it important to have a conjugate prior to the likelihood?

For example, if I define my prior distribution as Gaussian but I define my likelihood as multinomial, does this break any theoretical basis for variational inference?