Translumo is a real-time screen translation tool that captures and translates text from games, videos, and other on-screen content.
The tool solves the problem of translating hardcoded text and subtitles that cannot be extracted through conventional means. It works by combining multiple OCR engines simultaneously and using a machine learning model to score and select the most accurate recognition result. The captured text is then translated through integrated translation services. The approach prioritizes low latency and minimal system impact through several optimizations, making it suitable for real-time use in games.
Translumo is designed primarily for PC gamers who encounter untranslated or poorly translated content, though it works with any application displaying text on screen. It suits projects where text cannot be accessed programmatically and must be recognized visually. The tool supports multiple OCR engines including Windows OCR, Tesseract, and EasyOCR, with Windows OCR recommended for most users. Translation services include DeepL, Google Translate, Yandex Translate, and Naver Papago. Recognition covers English, Russian, Japanese, Chinese, and Korean, while translation extends to numerous additional languages. System requirements range from minimal for Windows OCR and Tesseract to substantial for EasyOCR, which requires an NVIDIA GPU with CUDA support and eight gigabytes of RAM.
The project maintains a substantial user base, with most open issues originating from adopters reporting real-world problems rather than from the core team. Issues and pull requests typically experience delays of weeks or longer before receiving initial responses. Work in the issue tracker centers on bug reports, enhancement requests, and general improvements.
