• Should be easier to install on Linux due to getting rid of the python dbus dependency (thanks to @KamWithK and @Manhhao for improving the Wayland mss shim)
• Added the option -swp/--screen_capture_wayland_persistence to persist the wayland screen capture selection/skip the portal popup. Running owocr with -swp=False once will reset the selection
• Fixed rare crashes with text reordering introduced in 1.26.5

Full Changelog: 1.26.7...1.26.8

1.26.7: fixed compatibility with older meikiocr

• Fixed text reordering not grouping furigana into the same paragraph if the furigana filter was off
• As I found a game where this was not ideal, with continuous screen capture the join_lines option is not implicitly enabled anymore
• Replaced meikiocr's unrecognized characters with spaces, and added support for getting text orientation for lines from meikiocr directly

What's Changed

• filter special character in meikiocr by @kamperemu in #78

Full Changelog: 1.26.5...1.26.7

1.26.5: made paragraph merging even stricter by requiring that all lines in a paragraph are merged as well

• Due to screen ai and Lens (with vertical text) splitting lines into multiple lines more often than I'd like to, merge_close_paragraphs is now on by default again. I have made it much stricter which should prevent a majority of glitches, but it might still be preferable for it to be off when processing manga
• Added a support_center_aligned_text option to support horizontal text that's center-aligned when reorder_text is enabled. It's off by default as it might cause some glitches/unwanted merging
• Restored the built-in manga ocr text processing which should bring the manga ocr segmented output in line with mokuro
• Text reordering has now basic support for RTL text

Full Changelog: 1.26.3...1.26.5

• Improve the text reordering code to separate paragraphs based on text alignment (thanks @rtr46 for the suggestion) rather than just line overlap/distance
• Improve the furigana filtering code to consider (approximated) text height rather than line width for vertical text
• When continuous screen capture and text reordering are both enabled, filter out extraneous lines earlier to prevent more mistakes during text reordering

Full Changelog: 1.26.2...1.26.3

• Made the part of the screen reordering code that merges paragraphs close to each other in the reading direction optional/off by default (I've only seen it being useful for the text of a single game with gaps in the middle of lines, but it caused glitches elsewhere sometimes that wrongly scrambled text)
• Made the --engines option accessible as a cli argument other than via the config file/configuration
• Restored the possibility for both engines to be local for two pass mode

Full Changelog: 1.26.1...1.26.2

• Updated builds to use Python 3.14 and updated readme to indicate 3.14 compatibility
• reading from a directory won’t ignore existing files unless you enable the new skip_existing_images setting (or pause on startup is enabled)
• write_to now requires directory names rather than the path of a single .txt file, and owocr writes single txt/json files there (if the image was a file it reuses the same filename, or it names the txt/json with a timestamp). The write_to directory can also match the same directory as read_from/read_from_secondary
• Updated the Azure implementation to use their new API since the old one was deprecated
• Updated NDL OCR Lite with the latest changes
  Full Changelog: 1.26.0...1.26.1

• implemented reading from OBS thanks to @kamperemu's work. This does not replace the built-in screen capture, it's an alternative for those who might prefer it. you need to enable websockets in the OBS settings and then configure it in the owocr config for it to work
• updated meikiocr to the latest changes from road_to_redemption, introducing a vertical text model
• implemented the new NDLOCR-Lite engine from the Japanese National Diet's Library (seems mostly suited to book scans and documents)
• added an optimisation for screen capture (and now OBS) where OCR won't be executed again if the screenshots are exactly the same. This should reduce CPU usage to at least some degree
• fixed several possible crashes and issues with the Windows exe

What's Changed

• Add support for reading from OBS Studio by @kamperemu in #65

New Contributors

• @kamperemu made their first contribution in #65

Full Changelog: 1.25.5...1.26.0

• Add auto downloading for screen AI files
• Fix terminal output from bat files on Windows 10

Fix meikiocr on Windows with latest pip packages/Windows 11 updates (?)

Prevent crashes in the logger while Screen AI is used