Translate manga from Japanese/Chinese to English using Gemini, or optionally with a Python server (YOLO+MangaOCR+mBART+Colorization)/in-browser CTD/Google.
These are versions of this script where the code was updated. Show all versions.
- Fix page freeze: window._MT state persistence across navigations, model cache per session, MT.processed prevents reprocessing same URL, force numThreads=1, rmv batch OCR (sequential ocrOneBox) + processQueue/queueImage with isConnected check
- Use browser-only mode by default.
- Fix senmanga.com: processAllImages -> processVisibleImage + img.complete check for stale race. (If error net::ERR_BLOCKED_BY_CLIENT appears in log, whitelist cdn.jsdelivr.net in your AdBlock.)- Fix DB detector noise-boxes: Filter out <6px or <1% img + Add mergeDbBoxes()- Fix Google Translate source: sl:'ja' -> 'auto' (auto-detect JP/ZH)- Remove manual Arrow Hotkey; by auto MutationObserver (150ms debounce)
- remove: grouping_box (CTD already proces Grouping.)
- fix: reload config on switch preset
- Fix: Bring back `function serverColorize`.
- Add WebGPU x4 speed-up (~7s): MangaOCR use WebGPU. (Dropped CPU (WASM backend)~1 min and Model `RT-DETR` unsupported by WebGPU, so replaced by `CTD` bubble-detection. Testing `db` text-detection for broader text coverage.) - /!\ WebGPU is tricky w/AdBlockˣDisable the extension if you encounter issues in browser mode.
- Add `.exe` provider compiled from the Python script using: `pyinstaller --onefile --name MangaServer --icon icon-.ico --collect-all torch --collect-all torchvision --collect-all cv2 --collect-all transformers --collect-all sentencepiece --collect-all safetensors --collect-data unidic_lite --collect-data fugashi --collect-data manga_ocr --hidden-import ultralytics --hidden-import manga_ocr --exclude-module torchaudio --exclude-module mamba_ssm --exclude-module torchdata --exclude-module expecttest --exclude-module matplotlib --exclude-module tkinter Server_Manga_Translator.py`
