Follow every lecture and talk on YouTube and Bilibili with smart-segmented bilingual subtitles, AI speech-to-text (ASR) for uncaptioned videos, timestamped chapter summaries, grounded video Q&A, and side-by-side web page translation. Use Lumi AI out of the box, or bring your own DeepSeek, OpenAI, Claude, or Paraformer key.
When a Bilibili video has no subtitles, AI transcribes the audio into timed bilingual subtitles.
Every lecture is split into timestamped chapters and key points; click one to jump there.
Search the subtitles for a phrase and jump to the second it was said.
English lectures on YouTube get re-segmented captions with a translation.
Auto-captions are rebuilt into full sentences and translated, not shown half a line at a time.
Ask about any concept; answers carry timestamps that take you back to the sentence.
Docs, GitHub and forum threads become side-by-side translations with links and inline code intact.
An hour-long talk becomes a structured reading draft that is far faster than watching.
Source and translation line up sentence by sentence, so you follow the original voice.
Click any subtitle line to hear it again instead of dragging the progress bar.
Set font size, color and position so long sessions stay easy on the eyes.
Export the subtitles as an SRT file to review offline or load into a player.
Read the key points and chapters first, then decide which part deserves a closer look.
Ask what the speaker thinks about something; the answer cites timestamps.
Save videos to your library and find the exact quote later by keyword.
Capture key frames into your notes, kept with their timestamps.
Paste a link and export SRT, VTT or TXT.
Remove rolling duplicates and merge fragments into readable paragraphs.
Convert between SRT, VTT and ASS, or shift the whole timeline.
Translate an SRT into 17 languages with the timing kept.
How large language models learn
大規模言語モデルの学習は二段階です。まずインターネット上のテキストで事前学習し、次に役立つアシスタントへと微調整します。
Large language models are trained in two stages: pre-training on internet text, then fine-tuning into a helpful assistant.
最初に `npm install` を一度実行し、読んでいる間は開発サーバーを起動したままにします。
Run `npm install` once, then keep the dev server open while you read.
Translations sit under every paragraph, or switch to translation only. Pick from four translation styles.
Bold text, links, and inline code stay as they are. Buttons, tables, iframes, and Shadow DOM text get translated too.
Toggle it from the floating button, a shortcut, or the right-click menu. Sites you keep translated stay translated next time.
Free Microsoft and Google engines work out of the box. Pro members get context-aware Lumi AI translation in multiple languages, or bring your own API key.
No captions? Transcribe first. Then refine the breaks and show a bilingual track. Stuck on a line? Click it and hear it again.
No new tab. Answers carry timestamps, so one click takes you back to the sentence.
Chapters, a readable article, and clickable timestamps. Replay a moment without dragging the scrubber.
Lumi stays in the side rail. Recommendations, danmaku, and the next-up teaser stop pulling you away.
After you sign in, history and the library remember where you were and what you asked.
All core features are free to try. Free accounts receive 200,000 Lumi AI tokens, while Pro includes 10,000,000 tokens each month. Bringing your own API key is always unmetered.
YouTube and Bilibili are supported now. Vimeo, Coursera, and other course platforms are next on the roadmap.
Yes. On any page, use the floating button, a keyboard shortcut, or the right-click menu to translate the whole page bilingually or as translation only. Free Microsoft and Google engines are available to everyone; translating pages with Lumi AI requires Pro, or you can use your own API key.
You can use Lumi AI with zero setup, or bring your own key for DeepSeek, OpenAI-compatible, Anthropic and Gemini models — pick whichever fits you in extension settings.
While signed out, progress stays on this browser. After you sign in, recent study and library items sync so you can continue on another device. We do not sell that data.
Yes. When the platform provides captions Lumi uses them; for videos with no captions at all, Lumi generates subtitles from the audio first, then translates, summarizes and chats over them.
Add to Chrome
Add to Edge
