Is There a Parallel Line of Research in Machine Learning?

In general, in order to get the disentangled per layer manipulation, you need a model where the input nois vector is fed not only into the first layer of the generator, but also all the subsequent layers. So i would say any model that has roughly this same structure would be potentially editable by our method. It's interest these techniques so far are just manipulating the outputs of gans o, fictional photos or images generated by these models. But certainly there is still some room to change the architecture in a way where we can get even more specific control of individual aspects in the image.