LocalSubs0.7.0

LocalSubs translates English subtitles from streaming video into Traditional Chinese in real time, keeping the processing on your computer instead of sending subtitle text to an external service. It preserves the original English dialogue while adding Taiwan-optimized Traditional Chinese subtitles, helping viewers follow English content they might otherwise miss. Translation uses a fine-tuned 0.6B local model designed for this task.

The Chrome extension is intended for HBO Max and works with Chrome, Chromium, and Microsoft Edge. Its Manifest V3 background service worker routes subtitle translation through either WebLLM, the default backend for new installations, or the optional native backend. WebLLM runs the model in the extension through WebGPU, is GPU-only, and requires a browser-exposed WebGPU adapter with shader-f16 support; unsupported hardware can use the native helper instead.

The native backend connects to localsubs_helper through Chrome Native Messaging and runs llama.cpp on a CPU or NVIDIA CUDA, with CUDA providing the lowest latency. New installations use WebLLM without the helper, while existing native-only setups retain that backend; switching backends unloads the inactive engine or resets the native connection without restarting the browser. Cached WebLLM artifacts use IndexedDB, and the content script can preheat a cached model but leaves preparation of an uncached model to the extension settings page.

Aiden1020’s open-source System Tools software, version 0.7.0, supports Windows 10 or later, Windows Server 2016 or later, and x64 systems, while the extension also supports macOS on Apple Silicon or Intel. The current release adds browser-based WebLLM inference and changes artifact caching to IndexedDB; its production model uses a 256-token context window, a 16-token output limit, and approximately 272 MB of q4f16_1 weights.