English | Русский
This version of Translumo has been updated to Version 1.5, focusing heavily on ONNX offline translation stability, CPU optimization, and ease of use.
- ONNX Language Setup Guide: Added a built-in, step-by-step UI guide inside the Settings menu to help users easily install and configure new language models for the ONNX Opus-MT translation engine.
- ONNX CPU Spiking Resolved: Fixed a severe issue where ONNX would aggressively consume 100% of all CPU cores. By enforcing
IntraOpNumThreads=1andInterOpNumThreads=1, ONNX now runs stably on a single thread, greatly reducing system load and eliminating CPU spiking during gameplay. - Infinite Loop & Garbage Text Fix: Resolved a critical greedy-search decoding bug where the
bad_words_ids(specifically the pad token54795) were not suppressed. This ensures the decoder properly terminates and produces clean text. - Language Switch Detection: Fixed a bug where ONNX model sessions were not properly disposed when users switched language pairs (e.g., from
en-idtoja-id), preventing stale translations. - Resilient Model Downloads: Improved the model downloader by using
.tmpfiles and atomic renaming. This prevents corrupted.onnxfiles if the application is closed or internet drops during a download.
This version of Translumo has been heavily upgraded to Version 1.4, bringing powerful new translation engines and critical crash fixes for offline usage.
- NVIDIA NIM Integration: Added support for NVIDIA NIM (NVIDIA Inference Microservices) as a translation engine, allowing blazing fast translations using NVIDIA's optimized AI containers.
- NVIDIA Riva Support: Integrated NVIDIA Riva for advanced AI-driven workflows.
- ONNX Offline Translation (Opus-MT): Integrated completely offline translation capabilities using ONNX CPU/NPU execution for HuggingFace
opus-mtmodels. The system automatically downloads the required models and dynamically patches Rust tokenizer incompatibilities on the fly!
- ONNX Tokenizer Rust Incompatibility Fix: Resolved a critical JSON parsing crash (
Tokenizers.DotNet.TokenizerException) when loading HuggingFace models by implementing an on-the-fly JSON patcher that removes incompatible JSnormalizerblocks without breaking the JSON structure. - Configuration Save Crash (Gamepad Hotkey): Fixed a major bug that caused the application to Force Close when exiting. The issue was traced to an unimplemented
GetHashCode()method inGamepadHotKeyInfo.csduring XML serialization, which has now been fully implemented with a stable hash calculation.
This version of Translumo has been customized and updated to Version 1.3. It includes exclusive features and major stability fixes that are not present in the original repository.
- AI Translator Integration: Added support for translating via Advanced AI Models (e.g., Gemini, ChatGPT/OpenAI compatible APIs) to achieve highly contextual and natural game translations.
- Local LibreTranslate Support: Fully integrated a completely offline, local LibreTranslate server support. Translumo will automatically manage the server startup and shutdown.
- Clear Chat Hotkey (
ALT+C): A newly implemented global hotkey that allows you to instantly clear the chat bubble/translation screen when it gets too cluttered.
- Proxy Error Fixes: Resolved persistent proxy errors that blocked connections to translators.
- LibreTranslate Lifecycle Manager: Fixed a bug where simply opening or minimizing the Settings menu would aggressively kill the LibreTranslate server. The server is now safely managed by a background singleton manager and only shuts down when you completely exit Translumo from the tray.
- Restored Anti Double-Click: Re-implemented the lock on the "Run LibreTranslate" button to prevent accidental spam-clicking that could cause duplicate server spawns.
- Dark Theme Removal: Systematically removed the highly unstable, experimental Dark Theme feature that was causing Translumo to force-close and crash on startup.
This project has a sibling called Lookupper — an on-screen dictionary for language learning. It is similar to Translumo but built for a different purpose. Lookupper is built to help you learn a language, not just depend on a translator forever.
Lookupper is my commercial project with a free version. If you find it useful and decide to grab the Pro version, you'll also be supporting the development of both Lookupper and Translumo.
Direct download link to the latest version:
Translumo_1.0.2.zip
After downloading, unzip the archive and run Translumo.exe.
Version 1.0.x includes many changes and improvements compared to versions 0.9.x. You can view the full list of updates on the Releases page.
-
High text recognition precision
Translumo allows combining multiple OCR engines simultaneously. It uses a machine learning model to score each OCR result and selects the best one.
-
Game oriented
Designed for real-time translation in PC games, but works anywhere on the screen with any application. -
Low latency
Several optimizations reduce system impact and minimize latency between text appearance and translation. -
Integrated modern OCR engines: Windows OCR (recommended), Tesseract 5.2 (legacy), EasyOCR (legacy)
-
Available translators: DeepL (recommended), Google Translate, Yandex Translate, Naver Papago.
-
Supported recognition languages: English, Russian, Japanese, Chinese (Simplified), Korean.
-
Supported translation languages: English, Russian, Japanese, Chinese (Simplified), Korean, French, Spanish, German, Portuguese, Italian, Vietnamese, Thai, Turkish, Arabic, Greek, Brazilian Portuguese, Polish, Belarusian, Persian, Indonesian, Bulgarian, Czech, Danish, Estonian, Finnish, Hungarian, Lithuanian, Latvian, Dutch, Romanian, Slovak, Slovenian, Swedish, Ukrainian.
- Windows 10 version 2004 (build 19041) or later, or Windows 11
- DirectX 11 compatible GPU
- 2 GB RAM
- NVIDIA GPU with CUDA SDK 11.8 support (GTX 750, 8xxM, 9xx series or newer)
- 8 GB RAM
- At least 5 GB of free storage space
- Open the Settings (Alt+G)
- Select languages: source language for OCR and translation language
- Select text recognition engines (see Usage Tips for recommended modes)
- Define the capture area: press Alt+Q and select an area on the screen
- Run translation (press ~)
- It is recommended to use WindowsOCR only.
Tesseract is old, slow, and produces many errors.
EasyOCR is even slower, requires significant resources (including a specific GPU), and often leads to bugs.
It’s probably better to remove all other OCR engines and keep only WindowsOCR, but they are still included in Translumo for historical reasons.
Reducing the capture area decreases the chance of picking up random letters from the background. Larger frames take longer to process.
Some translators may block clients sending many requests. Configure personal or shared IPv4 proxies (1-2 is usually enough) under Languages → Proxy tab. The app will alternate proxies to reduce requests from a single IP.
These modes are required for correct translation overlay display. If your game does not support them, use tools like Borderless Gaming.
Q: I get "Failed to capture screen" or nothing happens after translation starts
A: Ensure the target window is active. Restart Translumo or reopen the target window if needed.
Q: Borderless/windowed mode is set, but the translation window is under the game
A: With the game running and focused, press the hotkey (Alt+T by default) to hide and show the translation window.
Q: EasyOCR package download failed
A: Try reinstalling while connected to a VPN.
Q: Hotkeys don't work
A: Other applications may be intercepting hotkeys.
Q: Text detection failed (TesseractOCREngine)
A: Ensure the application path contains only Latin letters.
Visual Studio 2022 and .NET 8 SDK are required.
-
Clone the repository (the master branch always corresponds to the latest release):
git clone https://github.com/ramjke/Translumo.git
Note: During the build, binaries_extract.bat will automatically download and extract models and Python binaries (~400 MB) to the target output directory.
- Lookupper — on-screen dictionary and translator for language learning.
- ScreTran — simple screen translator.
- ScreenTranslator - screen capture, OCR and translation tool.