Lumi / Features
Bilibili learning, even without captions
Learn from Bilibili creator uploads, anime, and courses with native bilingual captions or AI speech-to-text (ASR). Start watching with the opening 5-minute preview segment while the rest of the video transcribes in the background.
Examples & specifications
| Item | Details |
|---|---|
| Video identity | Creator uploads (UGC): bvid + cid; Bangumi & courses (PGC): epId + cid (subject to standard platform access and audio availability). |
| Preview & durable queue | Managed ASR returns an opening Preview Segment of up to 5 minutes first, then processes subsequent 1-hour segments sequentially. |
| Shared transcript reuse | Once a full transcript passes coverage and Stable Audio Identity checks, Pro members can reuse it across accounts at 0 ASR duration cost. |
| Bring your own ASR | Supports Aliyun Paraformer / DashScope and Doubao Seed Flash ASR using your own provider account and billing. |
| Time | Source | Lumi translation |
|---|---|---|
| 16.49–24.39 s | I spent 10 years working for | 我花了十年时间为 |
| 24.39–28.78 s | Steve Jobs spent 10 years building Apple. 10 years | 史蒂夫·乔布斯花了十年时间打造苹果公司。整整十年 |
| 28.78–31.09 s | And he lost it all in a single moment | 而他在一瞬间失去了一切 |
Excerpt from the live product demo in its original English-to-Chinese language pair, illustrating how Lumi aligns and translates real cues. Bilibili · BV1H2Tu6EEzH · 2026-06-28
Workflow
- Open any Bilibili video, episode, or course and confirm the active part (P) and audio track.
- Use native captions when available, or launch managed ASR / your own speech provider when a video has none.
- Start learning as soon as the opening 5-minute preview segment is ready while background segments continue in order.
- Pick your target translation language for bilingual subtitles and click any timestamp to replay the original audio.
Questions & answers
- Can I generate bilingual subtitles for Bilibili videos that have no CC track?
- Yes. Lumi identifies Bilibili creator uploads by bvid + cid and Bangumi episodes or courses by epId + cid. When a video has no platform captions, you can run managed ASR or connect your own speech recognition service to transcribe the accessible audio track into timed source captions, then translate them into your target language. Audio access follows your normal Bilibili sign-in, membership, and regional permissions; for passages with heavy background music, overlapping speakers, or dense technical terms, simply click the timestamp to verify against the original audio.
- How does the 5-minute preview segment work? Do I have to wait for the whole video to finish?
- You don't have to wait for the entire video. In managed ASR, Lumi processes an opening Preview Segment of up to 5 minutes first and displays those subtitles as soon as recognition completes, while a durable cloud queue transcribes the remaining 1-hour segments sequentially in the background. First-segment turnaround depends on audio fetching and model inference speed, and starting a job reserves ASR duration quota for the full video length. Durable background queueing applies to managed ASR jobs, whereas BYOK speech services follow their respective API execution lifecycle.
- How does Shared ASR Transcript reuse work across accounts?
- A Shared ASR Transcript becomes eligible for reuse only after a complete transcription passes coverage and Stable Audio Identity verification. The original contributor can always use their own transcript, while other accounts with an active Pro membership can reuse verified transcripts immediately at 0 ASR duration quota cost. Only the timed subtitle text and non-sensitive provenance metadata are stored in the shared pool — never raw audio, API keys, signed URLs, partial segments, or personal edits. Matching is based on cryptographic audio identity rather than video titles.
- How do I connect my own Aliyun Paraformer or Doubao ASR key, and how is it billed?
- In extension speech settings, select Aliyun Paraformer / DashScope or Doubao Seed Flash ASR, paste your provider credentials, test the connection, and choose that provider when starting transcription on an uncaptioned video. Running BYOK Override consumes 0 Lumi quota; Alibaba Cloud or Volcengine bills your provider account directly according to their own pricing and rate limits. Complete BYOK transcriptions that pass Stable Audio Identity verification can also contribute to the shared transcript pool, while BYOK text translations stay strictly local.