This document proposes a method for continuous conversion of CT kernels using a switchable CycleGAN with adaptive instance normalization (AdaIN). It introduces soft and sharp CT kernels and the need for continuous conversion between them. The method uses a CycleGAN with a single generator and AdaIN to allow interpolation between kernel styles. Experimental results show the model can successfully interpolate between two kernels and among three kernels, outperforming alternative methods. The addition of a self-consistency loss further improves image quality.