PaneTrans

Free, private translation that runs 100% on your device: live YouTube subtitles, bilingual pages, and any screen region.

NEW IN v1.0 • Free for everyone — every feature and every model, no account, no license key. • 100% on-device — the cloud tier is retired; your text never leaves your computer. • Live subtitles and regions get priority while a page translates; faster page translation; better long paragraphs. PaneTrans translates the web on your own device. LIVE REGION TRANSLATION Drag a rectangle over a live chat, video subtitles, a stock ticker or a comment feed. A glass popup shows the original and the translation side by side and re-translates the moment the text changes — like clipping a real-time subtitle strip onto any slice of the page. • Persistent: the popup follows the region until you close it. • Reads live page text, and automatically falls back to OCR for video frames, images and canvas. • Caption-bar mode: subtitle-shaped regions render like a second subtitle track under the video. • Monitor several regions at once, each in its own draggable popup. YOUTUBE BILINGUAL SUBTITLES Turn it on and every captioned video shows the translation together with the original, with adjustable size. BILINGUAL PAGE TRANSLATION Press Alt+T (or use the popup) to insert the translation under each paragraph, or switch to "replace" mode. Hover a paragraph and tap Alt to translate just that one. Set sites to translate automatically. RIGHT-CLICK Select text and right-click to translate it instantly, or right-click anywhere to drag-select a region. ON-DEVICE MODELS — ALL FREE • OPUS-MT (default): small and real-time on WebGPU. • NLLB-200 / M2M-100: one model for many language pairs. • Qwen 3 (0.6B) / Gemma 3 (1B): on-device language models for more natural phrasing (WebGPU required, not real-time). English, Chinese, Japanese, Korean, French, German, Spanish, Russian, Portuguese, Arabic. PRIVACY No account, no ads, no analytics, no telemetry, no servers. Translation and OCR run entirely in your browser. The only network requests download model and OCR files on first use (Hugging Face, jsDelivr, tessdata); they contain none of your text. After that it works offline.