To start with, [github.com] https://github.com/msxie92/MangaInpainting has been around for more than a year. The author did not release its training part so its ability is limited.

However, this model is quite "modularized", as it is described in paper:
[smms.app] (IMG:[s2.loli.net] https://s2.loli.net/2022/12/15/2djQ8kLEOW1HFDw.png)

I think we can use it with BidirectionTranslation by the same author to extend it:
[smms.app] (IMG:[s2.loli.net] https://s2.loli.net/2022/12/15/NromyqXHYsDhWFZ.png)

I've cleaned the code of ScreenVAE(a step of "preprocess") and BidirectionalTranslation so that they can be used as python modules: [github.com] https://github.com/TsXor/ScreenStyle
But that was too much effort for a python beginner like me and I hope someone can continue it by cleaning the code of Appearance Synthesis part and try my idea out. (IMG:[invalid] style_emoticons/default/unsure.gif)