Hi all,
A new Release v4.1.0 for Crow Translate was published.
Recommended download source KDE - Experience Freedom! to save our Gitlab from heavy load.
Gitlab Assets:
Release notes:
Crow Translate 4.1.0.
New:
- LocalAI translation backend: translate through any Ollama-, LM Studio-, FastFlowLM-, OpenAI- or Anthropic-compatible server, with model discovery and configurable prompts
- Vision-model OCR: a second OCR engine that transcribes images with a vision model, selectable alongside Tesseract in settings. I would still recommend Tesseract, unless you really need the 10% accuracy gained from the 100x larger glm-ocr
- Modular OCR: screen capture, image drop/paste and the D-Bus interface all route through whichever OCR engine is active
- Open an image for OCR by dropping, pasting or opening a file, with a preview overlay
- Status strip in the main and popup windows showing what is running (capture, recognition, detection, translation, speech) and what failed
- LibreTranslate API key support for Mozhi instances, with a toggle between the Mozhi-proxied and direct LibreTranslate API shapes
- Report Bug action in the tray menu, pre-filling version and OS
- WITH_TTS build option to compile out the text-to-speech engines and their UI
Fixed:
- Capture area on Windows with multiple monitors at different scaling
- Missing application icon on Windows and macOS; icons now load through KIconThemes with a theme fallback
- Recovery after an unclean shutdown, which could leave the application unable to start
- The translate-selection shortcut on Wayland
- Hebrew on the Yandex and DuckDuckGo engines
- The autostart desktop file name
- Translation failures are now shown instead of being swallowed, and their messages are translatable
- Various typos
Accessibility and i18n:
- Icon-only buttons throughout the main window, popup window, settings dialog and languages dialog carry accessible names
- Label and control pairs are associated for screen readers and keyboard navigation
- User-facing error messages from every translation backend are translatable
Development:
- Regression test suite covering the translation providers, OCR engines, settings, status model and window behaviour
- ONNX Runtime is found from craft installs, builds against 1.27, and tolerates newer GCC. No longer using the binary build from Microsoft, telemetry neither enabled nor built.
Checksums
70cab22a089aa976627fb0f04ea2bf2e918a138c36ff0337d4a7709ea5462fa3 crow-translate-4.1.0.tar.gz
554b47f444f8ce23d51c7a46c63d2e80ef3851547290d4ae4dc19efcd9deef88 crow-translate-4.1.0-windows-cl-msvc2022-x86_64.exe
2e169e9918db67873a794d80563721e21d2c531ce0f90a6cff3a8c784ed1a571 crow-translate-4.1.0-windows-cl-msvc2022-x86_64.7z
fb9fe8cc6885d25fe330060471c7a9e3f4a038fc80966ebd39f5125c239f2fb4 crow-translate-4.1.0-linux-gcc-x86_64.AppImage
Special thanks to Oleksandr Mikriukov (Oleksandr Mikriukov · GitLab) who put in most of the seminal work for the AI backends.
Coming up next: Full plugin architecture, alternative frontends, scriptable translation provider
Regards,
The Crow Translate Team