Libre Manga Translator by Mrdhnto

Libre Manga Translator — in-place manga translation, you choose where your data goes.

Hover any manga page image to show the Translate pill. Detection runs on-device, you confirm boxes, then it translates and repaints the page. After translation the pill becomes an action hub: edit text, adjust boxes, toggle Original/Translated, export JPG.

1. Local (default) — Fully on-device: bubble detection + OCR + Qwen3.5-4B / Tiny Aya Global. Works offline after one-time download. No account, no uploads.
2. Gemini (opt-in, your key) — Browser-to-Google direct, OCR + translation in one call. For low-GPU machines.
3. API Mode (opt-in) — Ollama, LM Studio, or any OpenAI-compatible /v1/chat/completions endpoint. Only extracted text is sent, never images.

• Detection: RT-DETR Bubble Detector (default, bubbles + free text) and Comic Text Detector with pixel-mask seeding, sorted in manga reading order.
• OCR + Script Gate: PaddleOCR multilingual (11 packs, auto-switched), PP-OCRv6 Manga lightweight, Manga-OCR vertical specialist. On-device gate holds back SFX / wrong-language regions with one-click Translate anyway.
• Inpainting: Fast ladder (planar fill → denoise → Telea) and Quality mode (LaMa neural redraw first, Fast fallback per region). No ghost boxes.
• Reader: Auto-translate on scroll (one page at a time), drag / resize / add / delete / undo bubble editor, series title + glossary + context across pages, site adapters + AI rule generator, custom TTF/OTF/WOFF fonts.
• Control: Model Storage manager (list / download / update-check / delete), master WebGPU switch + per-area LLM / inpaint / OCR overrides with WASM fallback, local-only debug logs, Obsidian Noir dark + light theme.
• Security: Zero telemetry, machine-local API keys, SSRF-gated image proxy, strict schema validation, SHA-256 model verification.

No account. No analytics. Local models download once (~0.8–4GB depending on LLM pick), then run offline.

Please support the official releases of manga you read.