MashiroSaber03/Saber-Translator

What it solves

Saber-Translator is a comprehensive tool for translating and managing manga. It automates the tedious process of converting foreign-language manga into a reader's native language by handling text detection, OCR, translation, and image inpainting, while also providing tools to organize libraries and analyze story content.

How it works

The project implements a multi-stage pipeline for translation and a RAG-based system for content analysis:

• Translation Pipeline: The workflow follows a sequence of importing images/PDFs $\rightarrow$ detecting text bubbles (using DBNet, CTD, or YOLO) $\rightarrow$ OCR recognition (via MangaOCR, PaddleOCR, etc.) $\rightarrow$ AI translation (via DeepSeek, Gemini, Ollama, etc.) $\rightarrow$ image inpainting to remove original text (using LAMA) $\rightarrow$ rendering the translated text back into the image.
• High-Quality Mode: An experimental mode that uses multimodal models to analyze multiple images simultaneously, providing better contextual awareness for more coherent translations.
• Manga Insight: An AI engine that uses a four-tier analysis architecture (Page $\rightarrow$ Paragraph $\rightarrow$ Chapter $\rightarrow$ Book) to summarize plots and identify characters. It employs Retrieval-Augmented Generation (RAG) with vector search and reranking to allow users to ask intelligent questions about the manga's plot.
• Management: A built-in bookshelf system for organizing books, chapters, and tags, paired with a dedicated reader interface.

Who it’s for

It is designed for manga readers and translators who want to automate the translation of comics and maintain a structured digital library of their translated works.

Highlights

• End-to-end Workflow: Integrates everything from raw image import to final rendered export (PNG, ZIP, PDF, CBZ).
• Advanced Inpainting: Uses LAMA intelligent repair to cleanly remove original text from bubbles.
• Context-Aware Translation: Multimodal support for translating across multiple pages to maintain story flow.
• Intelligent Analysis: RAG-powered Q&A, character mapping, and automated plot timelines.
• Flexible Editing: A manual editor for fine-tuning text placement, fonts, and colors.