On-screen OCR and translation for privacy-conscious Windows users
Experience instant on-screen capture that recognizes unreadable text and renders translations as an overlay. PingYi, created by PingYi contributors, is a Windows utilities tool that combines local OCR and optional LLM translation. It runs PaddleOCR locally for offline text recognition, connects to local LLMs through llama.cpp or Ollama for context-aware translation, and fits gamers, researchers, and international users who need translations without switching apps.
What does the tool do when text cannot be selected?
PingYi captures a user-selected screen region via a global hotkey and runs OCR to extract text, then displays a translation overlay. The app targets cases where text is embedded in images, hardcoded subtitles, or game dialogue, performing recognition and translation in one workflow. The interface supports English and Chinese, and the history feature records recognized strings for later reference.
Does it affect system performance during recognition?
Moreover, the app is designed for local processing: PaddleOCR performs offline recognition and local LLMs handle translation, which confines CPU and memory use to the machine. The underlying Avalonia UI keeps the client lightweight on Windows desktop environments. Scan and translate operations run on demand after a capture, so background system load is limited to idle resource use rather than continuous scanning.
Is it safe and private to use on a desktop?
In addition, PingYi uses an offline-first architecture that runs OCR and LLM models locally when configured, which reduces network exposure. The tool still supports online services like Google Cloud Vision, but local model options let users avoid cloud-based processing. The modular design permits swapping OCR engines, so users can keep recognition inside their environment for privacy-sensitive material.
A focused pick for users who need private, on-screen translations
PingYi suits users who encounter non-selectable text and prefer local processing: it binds capture, OCR, and translation into a single interaction and supports local LLMs for context-aware output. The main caveat is dependency on model setup for offline LLMs, which requires additional configuration and local resources. For anyone prioritizing privacy and a single-step workflow, the app performs the core task reliably.
