ZeroTang05/VideoAdGuard
哔哩哔哩浏览器插件:基于大语言模型,对B站视频中的植入广告进行检测。一键跳过视频中的植入/口播广告。
What it solves
VideoAdGuard is a browser extension designed to detect and skip integrated advertisements (sponsored segments) within Bilibili videos. It helps users avoid promotional content by automatically identifying the specific time intervals where ads appear.
How it works
The tool analyzes video content using a combination of text and audio analysis powered by Large Language Models (LLMs). The process follows these steps:
- Content Extraction: It retrieves video subtitles, titles, and pinned comments. If no subtitles are available, it downloads the audio stream and uses the Groq Whisper API for speech-to-text conversion.
- Cache Check: It first checks a local 24-hour cache and a community-driven cloud cache to see if the ad segments for that video have already been identified.
- AI Analysis: If no cached result is found, the subtitles or transcribed audio are sent to a configured LLM (such as DeepSeek, OpenAI, or local Ollama models) to determine the exact start and end times of the advertisement.
- User Interface: Once identified, the extension adds a "Skip Ad" button to the player and marks the ad segments in red on the video progress bar.
Who it’s for
Bilibili viewers who want to save time and avoid sponsored segments in videos, as well as users who prefer using their own AI API keys for content filtering.
Highlights
- Multi-Model Support: Compatible with OpenAI, Anthropic, and local models via Ollama, LM Studio, or vLLM.
- Audio Recognition: Can detect ads in videos without subtitles using the Whisper model via Groq.
- Community Sharing: Features a cloud cache that allows users to share and correct ad timestamps, improving accuracy for everyone.
- Interactive Marking: Users can manually adjust the start and end times of detected ads on the progress bar and submit corrections to the cloud.
- Automatic Skipping: Option to automatically jump over identified ad segments without manual clicking.
Related
- Project
- Project
- Project
- Project