bilive

Automatically monitors and records Bilibili live streams and danmaku (including paid comments, gifts, etc.), converts danmaku to match video resolution, uses speech recognition for subtitles and renders them into videos, splits exciting clips based on danmaku density, generates fun titles via large video understanding models, auto-uploads videos and clips to Bilibili, compatible with no-GPU versions, works with extremely low-spec servers and PCs.

• 🚀 Fast speed: Uses pipeline processing for videos, under ideal conditions, recordings are only half an hour behind the live stream, so they can be uploaded before the stream ends. Currently the fastest known Bilibili live recording version!
• 🎥 Multi-room support: Record videos and danmaku files from multiple live rooms simultaneously (includes regular danmaku, paid danmaku, and gift/crew membership info).
• 💾 Low storage footprint: Automatically deletes locally uploaded videos to maximize space savings.
• 📦 Template-based: No complex configuration required, works out of the box. (🎉 NEW) Automatically fetches relevant trending tags via Bilibili's search suggestion API.
• 🔍 Automatic segment detection and merging: Automatically detects and merges split video streams caused by network issues or live stream connections into complete videos.
• 🖥️ Auto danmaku rendering: Automatically converts XML to ASS danmaku files, renders them into videos to create danmaku-enabled videos, and uploads them automatically.
• ⚡ Extremely low hardware requirements: No GPU needed, only a basic single-core CPU with minimal RAM is sufficient to complete recording, danmaku rendering, uploading and all other processes. No minimum spec requirement, even PCs or servers from 10 years ago can run it!
• (🎉 NEW) Auto subtitle rendering (Nvidia GPU required for this feature): Uses OpenAI's open-source Whisper model to automatically recognize speech in videos, convert it to subtitles and render them into the video.
• (🎉 NEW) Auto clip upload: Calculates high-energy clips based on danmaku density, uses the multimodal large video understanding model GLM-4V-PLUS to automatically generate engaging clip titles and content, and uploads them automatically.