Live screen translator that makes OCR engines compete

What it does Translumo is a Windows desktop tool that captures a screen region, runs OCR, and overlays the translated text live. Built for PC games with hardcoded subtitles or foreign UI, it works with any on-screen text. Several OCR backends and translation services sit behind a single interface.

The interesting bit Instead of trusting one OCR engine, Translumo runs multiple in parallel and uses a machine-learning model to score their outputs, picking the best. The README admits most bundled engines are legacy baggage—Windows OCR is the only one the author recommends—so the ensemble is a hedge against garbled text, not a marketing bullet.

Key highlights

• Runs multiple OCR engines simultaneously and votes on the best result via an internal ML scorer.
• Optimized for low-latency overlay in games; supports borderless and windowed modes.
• Integrates DeepL, Google Translate, Yandex, and Naver Papago as translation backends.
• Recognizes English, Russian, Japanese, Chinese (Simplified), and Korean source text.
• Ships as a standalone Windows executable; proxy rotation is included to reduce translation-service rate-limiting.

Caveats

• The author explicitly recommends using only Windows OCR; Tesseract and EasyOCR are described as slow, error-prone, and resource-hungry legacy options.
• EasyOCR demands an NVIDIA GPU with CUDA 11.8, 8 GB RAM, and 5 GB of disk space.
• Translation overlays require borderless or windowed game modes; true fullscreen is unsupported without third-party workarounds.

Verdict Grab it if you need live translations for untranslated games or videos with burned-in subtitles. Skip it if you are on Linux or macOS—it is strictly Windows 10/11.