Posted July 13, 2025 by Godnoken
Note 1: Remember that you have to download each update manually. GameTranslate does not currently have any auto-update support!
One full week behind schedule - 0.4.8 is finally out.
I've been fully struggling to maintain a good work/life balance these past weeks, which more or less culminated last week. Hence the major delay and radio silence. I do apologize for that and hope that I can better my circumstances for the next few weeks so that GameTranslate gets the attention it needs.
The automatic mode often suffered from paragraphs having their texts overflowing into each other, making it more often than not unreadable to select areas with many paragraphs. This was worked on for 0.4.8 and should now no longer be occurring. Please do leave a comment if you still have this problem.
I have also completely rewritten the text line merging algorithm. The old one tried to fix too many edge cases, which in the end just broke more than it fixed. The new algorithm is simpler and way less prone to breaking up text, that in turn could make translations intelligible. There are now also two user-defined x and y thresholds found in the OCR tab when you have RapidOCR selected. If you experience paragraphs being split up, increase the y threshold, and vice versa if texts that should be separated are merged into paragraphs. The x threshold should be increased if you find text lines being separated horizontally when they shouldn't be.
Just recently I noticed that selecting very wide areas using RapidOCR would decrease the text output to an extent that it basically did not capture any text at all. This ought to work much better in 0.4.8, but due to me at the moment being limited to a 1080p screen for testing, I am not entirely sure if this still exists for 2K and 4K monitors. If you have one and experience issues capturing text using RapidOCR, please let me know!
Have you had "Translation model failed" displayed in your translation box when using the app? Good news, there are now user-defined settings available in the 'Translation' tab when the 'Internal' engine is selected. You'll now see CPU Threads, Workspace and Cache Translations. To prevent the model from crashing and failing, make sure the cache is unchecked, and set the workspace and CPU threads to a lower value. Be aware that this will make translations slower.
Lastly, but not least - Three new RapidOCR (PaddleOCR) models are bundled with this release.
They are Korean, Latin and Cyrillic.
Not all Latin languages are made available to use the Latin model due to the quality simply not being good enough according to the testing I've done. The same goes for the Cyrillic languages. I hope new and more focused models will keep coming out since RapidOCR works so much better than Tesseract in more or less all cases.
The list of languages made available can be found in the changelog below.
As always, if you have the time - please do leave feedback and voice your opinion on what you'd like to be changed, improved or added in the future. It is a huge motivating factor for me to put in that extra bit of work to make the app live up to its potential.