Hmm, interesting. This is definitely not something I thought of when designing the image conversion feature - support for multi-color character sets didn't exist back then. It seems like what the program should do here, to get the result you want, is reduce the image to the *character set's* palette instead. I'll put this on the TODO list (trello), but I'm not sure how much work is involved and can't offer any ETA on when it might happen. Thanks for raising this issue, sorry I didn't have a ready answer.