StencilTorch: An Iterative and User-Guided Framework for Anime Lineart Colorization.

Automatic lineart colorization is a challenging task for Computer Vision. Contrary to grayscale images, linearts lack semantic information such as shading and texture, making the task even more difficult. Modern approaches train a Generative Adversarial Network (GAN) to generate illustrations from user inputs such as color hints. While such approaches can generate high-quality outputs in real-time, the user only interacts with the pipeline once at the beginning of the process. This paper presents StencilTorch, an interactive and user-guided framework for anime lineart colorization motivated by digital artist workflows. StencilTorch generates illustrations from a given lineart, color hints, and a mask allowing for iterative workflows where the output of the first pass becomes the input of a second. Our method improves previous work on both objective and subjective evaluations.