Continuous user signal provides rich feedback for navigating latent spaces
Traditionally, when having to find a local optimum in a high-dimensional latent space, such as when editing a 3D model or an image, users would sequentially select from several alternatives (e.g. 2GC, 4GC), informing the search process. However, a continuous signal (e.g. through a slider) offers faster convergence of the search process and better results, even with the slight time overhead.