Ah! I see. What you're speaking of here is not a new problem for 0.5.0. It has always been there. I've tried to fix it in the past, but did not succeed at that point in time. I know how to fix it in two different ways in theory though and it will be worked on very soon as this is crucial for many games.
The problem right now is that I allow new translation requests as soon as the image changes by even just one pixel. This is the most performant check I can make, but obviously it has massive drawbacks.
When you tried the manga, you saw text jumping around even if you completely let go of keyboard and mouse? If so, that's definitely not intended and something I'd have to look into.
About recognition.. yes, sadly, OCR is very complex and I ought to give the users a few more configurable values so it fits their use cases better. There's never going to be a one size fits all solution for this, no matter how much I'd like it to be as simple as possible for everyone. :/
Thank you very much for your feedback!

























