MMLab@NTU

In this work, we successfully achieve highly controllable multi-modal image colorization. Our method is built on Stable Diffusion with new designs such as stroke control and content-guided deformable autoencoder. These designs offer our method realistic outputs along with various applications, including text/stroke/exemplar-based colorization, hybrid control colorization, recolorization, iterative editing and the pioneering implementation of local region colorization. Our approach provides a more diverse range of colors and high user interactivity.