How to bulk-analyze bilibili subtitles, chapters and covers
Analyze bilibili transcripts, subtitles and covers on your own Windows PC: chapter markers, caption QC, image lens and video studio, all running offline.
Skills in this bundle
Video Content Studio
Content video production copilot (task-mode x top_n; LLM-orchestrated). (Action Co-pilot: generates next-best-actions (actions.json), follow-up tasks and calendar reminders on top of the analysis — what to act on today.)
Chapter Marker
Segment a timestamped transcript into chapter markers by pause gaps. (Action Co-pilot: generates next-best-actions (actions.json), follow-up tasks and calendar reminders on top of the analysis — what to act on today.)
Caption Inspector
QC subtitles: reading speed (CPS), timeline overlaps, line-count/length limits. (Action Co-pilot: generates next-best-actions (actions.json), follow-up tasks and calendar reminders on top of the analysis — what to act on today.)
UI Studio
UI Studio process copilot (task-mode x top_n; LLM-orchestrated). (Action Co-pilot: generates next-best-actions (actions.json), follow-up tasks and calendar reminders on top of the analysis — what to act on today.)
Visual Design Studio
Visual Design Studio process copilot (task-mode x top_n; LLM-orchestrated). (Action Co-pilot: generates next-best-actions (actions.json), follow-up tasks and calendar reminders on top of the analysis — what to act on today.)
Image Lens
Media task skill: Image Lens (sub-variant mode x top_n; LLM-orchestrated). (Action Co-pilot: generates next-best-actions (actions.json), follow-up tasks and calendar reminders on top of the analysis — what to act on today.)
Put your exported transcripts, SRT files, cover images and script drafts in one folder, then run Video Content Studio, Chapter Marker, Caption Inspector and Image Lens from Nexus Runner: you get content breakdowns, chapter timecodes, a caption defect list and cover composition analysis as structured result files. Every step executes on your own Windows machine, so the material never leaves it.
- Local input: transcripts, subtitles and covers sit in one folder on your PC, and the skills read those files directly.
- Four-step chain: Video Content Studio breaks down the content, Chapter Marker cuts the timeline, Caption Inspector flags defects, Image Lens reads the cover.
- Deterministic where it matters: chapters come from pause gaps in the transcript, caption findings from CPS, overlap and line rules.
- Visual pass included: Visual Design Studio covers palette and typography, UI Studio reviews cover templates and collection-page layout.
- Cost is arithmetic: 12 Nex for a full four-step pass on one video, against a free tier of 200 Nex per month, reset at the start of each month.
What bulk bilibili content analysis means here
Bulk analysis here means handing a batch of already-exported assets (timestamped transcripts, SRT subtitles, cover images, script drafts) to a fixed set of skills that apply the same criteria to every item, so the results line up and can be compared, instead of a person opening each video one at a time. Each skill does one defined job: Video Content Studio produces content breakdowns by mode (script writing, transcription, video analysis, adaptation, production planning) times top_n; Chapter Marker segments a timestamped transcript into chapter markers by pause gaps; Caption Inspector checks reading speed (CPS), timeline overlaps and line-count or line-length limits; Image Lens reads covers by composition, subject and style mode; Visual Design Studio handles icons, palette and typography; UI Studio produces wireframes and design reviews. All of them run inside Nexus Runner on your machine: files in, structured result files out.
From a folder of assets to structured results
Three ways to do this
| MCP Nexus | Manual work | Generic cloud AI | |
|---|---|---|---|
| Where files live | Reads transcripts, subtitles and covers from a local folder; processing and results stay on the machine | Assets scattered across personal drives and shared folders, moved around by hand | Transcripts, subtitles and covers have to be uploaded to the provider |
| Batch handling | One batch, one set of modes and top_n values, results with identical structure | Opened and noted one by one, with criteria that shift between people | Either write your own API scripts or paste items into a chat one at a time |
| Chapter and caption checks | Chapters from pause gaps; captions judged by CPS, overlap and line rules, with gaps left empty | Dragging the timeline and counting characters by eye | Left to the model, so the same subtitle file can come back scored differently twice |
| Output shape | Structured result files you can archive, diff and pass to the next step | Loose spreadsheets, notes and chat logs | A prose answer that still has to be reorganized into a table by hand |
| Cost model | Per-run Nex: 12 Nex for a full four-step pass on one video, against 200 free Nex per month | Mostly staff hours, growing linearly with output | Billed on the provider's own terms, separate from the local workflow |
Real use cases
FAQ
Do I have to upload anything to the cloud to analyze bilibili videos?
No. All six skills run inside Nexus Runner on your own machine, reading transcripts, subtitles and covers from a local folder, and both the processing and the results stay on that computer.
Can it log into a bilibili account and pull the data automatically?
No. Nexus does not drive the screen and does not sign into or operate third-party site accounts on your behalf; you export the material from the creator dashboard as text, subtitle or image files and hand those to the skills.
How many videos does the free tier cover?
The free tier is 200 Nex per month, reset at the start of each month. A full pass on one video costs 12 Nex (Video Content Studio 1, Chapter Marker 3, Caption Inspector 3, Image Lens 5), so roughly 16 videos; heavier use fits Pro at CNY 49 per month with 3000 Nex or Enterprise at CNY 299 per month with 50000 Nex.
Are the chapter splits and caption findings just made up by a model?
No. Chapter Marker and Caption Inspector run on local deterministic engines: the first segments by pause gaps in the transcript timestamps, the second applies reading speed (CPS), timeline overlap and line-count or line-length rules, and anything the rules cannot determine is left empty rather than filled in.
Does it work on a Mac, and which version do I need?
Windows 10/11 x64 only, with Nexus Runner 1.0.10.