how do I generate images from the w space?

the generate_images script samples latents from the z space. the script shows how varying each z_i at a time changes one aspect of the face. this means that the dimensions are interpretable and somewhat disentangled even in Z? 
the main disentanglement results are based on the w latent space tho. is there a way to generate the faces directly from the w latents instead of z?

thanks!!