Bili Mindmap

Turn a Bilibili video into a mind map that feels closer to something a human actually organized.

Recommended Flow

• - Python scripts collect video details, subtitles, AI summary, comments, and ASR fallback when needed.
• The host platform's injected model reads the prepared context and writes a high-quality outline.md.
• Python renders outline.md into an .xmind file.

Preconditions

• - bili must be installed and available.
• If audio fallback is needed, bilibili-cli[audio] should be installed.
• If cloud ASR is used on Windows, the Aliyun config file should already exist.
• If local ASR is preferred on Linux or macOS, make sure the Parakeet endpoint is running.

Core Constraints

• - Prefer subtitles first. Only fall back to ASR when subtitles are unavailable.
• Login check is mandatory: run bili status before bili login.
• The main way to produce outline.md should be the host model, not the local rule-based script.
• The main structure should come from subtitles or ASR. Comments and the site AI summary are supplemental only.
• Do not mechanically copy spoken transcript text. Merge themes, compress phrasing, and organize by logic.
• If information is weak or incomplete, mark it explicitly instead of inventing facts.

Main Workflow

1. 1. Accept either a full video URL or a BV id.
2. Run bili status to check login.
3. If needed, run bili login and wait for the user to scan.
4. Run python scripts/prepare_bili_context.py --source <video-url-or-bv> --login-if-needed --transcribe-if-needed.
5. Read the generated files: context.md, host_outline_prompt.md, manifest.json, video_details.json, subtitles.txt, ai_summary.txt, and comments.txt.
6. Feed host_outline_prompt.md to the host platform model and let it write outline.md. Only use scripts/generate_outline.py when the host model path is unavailable.
7. Run python scripts/render_xmind.py --outline <output-dir/outline.md> --output <output-dir/result.xmind>.
8. Tell the user where the .xmind file was written and which sources were most important.

One-Command Workflow

INLINECODE24 now supports two workflows:

• - --workflow host: recommended quality path. Collects context first, then waits for a host-generated outline.md.
• INLINECODE27: fallback path. Uses scripts/generate_outline.py locally.

Recommended command:

CODEBLOCK0

On the first run, if outline.md does not exist yet, the script will stop after context preparation and print:

• - the context.md path
• the host_outline_prompt.md path
• the expected outline.md path

After the host model writes outline.md, run the same command again and it will render the .xmind file.

Fallback Workflow

When the host model cannot be used, fall back to the local outline generator:

CODEBLOCK1

This is only a fallback. It is usually lower quality than the host-model result.

Collection Strategy

Collect information in this order:

1. 1. bili video <source> for video details
2. INLINECODE36 for subtitles
3. INLINECODE37 for the site AI summary
4. INLINECODE38 for hot comments
5. If subtitles are unavailable:

Output Requirements

• - Use the video title as the root topic.
• Keep subtitles or ASR as the main evidence.
• Prefer abstraction and synthesis over transcript copying.
• Mark uncertainty explicitly.
• The final artifacts should include both outline.md and .xmind.

Important Files

• - scripts/prepare_bili_context.py: login checks, content collection, ASR fallback, and generation of context.md plus INLINECODE46
• INLINECODE47: local fallback outline generator
• INLINECODE48: pure Python XMind renderer
• INLINECODE49: one-command entry point with host and local workflows
• INLINECODE52: structure template for the final outline
• INLINECODE53: quality and behavior rules for the host model path
• INLINECODE54: bundled Aliyun file transcription implementation