Video Translator by Tasknetic

Video Translator reads subtitles that are burned into the picture of a YouTube video — the kind that are part of the image itself, not YouTube's own closed captions — and paints a translation over the video as it plays.

Everything happens on your device. Frames never leave your computer, the recognized text never leaves your computer, and there is no cloud translation service, no account, and no API key.

HOW IT WORKS

Video Translator samples the video's subtitle band, recognizes the text with an on-device OCR engine, translates it locally with the Bergamot neural translation engine (the same engine family Firefox uses for its own built-in page translation), and overlays the result in sync with playback.

FEATURES

• Real-time translation of hardcoded subtitles on YouTube
• Fully on-device — nothing you watch or read is transmitted anywhere
• 18 translation target languages: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Japanese, Korean, Chinese (Simplified and Traditional), Arabic, Hindi, Turkish, Vietnamese, Thai, Indonesian
• Automatic detection of the subtitle's language, or pick the source language yourself for the most reliable recognition
• Adjustable overlay size, opacity and position
• Adjustable capture region, so OCR reads exactly the band your subtitles sit in
• A diagnostics panel showing recognition latency, cache hit-rate and sampling rate

SUBTITLE LANGUAGES THAT CAN BE READ

The OCR models are bundled with the add-on rather than downloaded, so the set of subtitle languages that can be read is fixed: English, Chinese (Simplified) and Chinese (Traditional), plus Japanese, Spanish, French, German, Italian, Portuguese and Dutch on the higher-quality OCR setting. Languages that cannot be read are shown but disabled in the picker rather than silently failing.

Translation targets are separate and cover all 18 languages listed above.

Auto-detect works best on the default OCR setting. On the lightest OCR setting the
reader falls back to the English model, so set the subtitle language explicitly if
you are watching Chinese subtitles there.

FIRST RUN

The first time you translate into a given language, the add-on downloads that language pair's translation model (roughly 15–20 MB) from Mozilla's translation model storage and caches it. Progress is shown in the popup. After that the pair works offline.

REQUIREMENTS

Firefox 140 or newer, on desktop. Works on youtube.com only.

WHAT IT CANNOT DO

DRM-protected videos cannot be captured — their frames are unreadable to any extension, and the add-on says so rather than failing silently. Videos with no burned-in subtitles produce no overlay; YouTube's own closed captions are DOM text and are not what this add-on reads.