Download YouTube videos, extract frames, and caption them with a vision model.
npx skills add https://github.com/ahostbr/dotcodex --skill video-lens
Please help me install this Agent Skill. Skill: video-lens Source: https://github.com/ahostbr/dotcodex/tree/main/skills/video-lens Command: npx skills add https://github.com/ahostbr/dotcodex --skill video-lens
Extracting insights from long-form YouTube videos is tedious; video-lens automates downloading, frame-by-frame extraction, and captioning/indexing to create searchable video metadata and visual summaries.
Run video-lens with a YouTube URL to download the video, extract frames at a chosen interval, and generate a JSON index plus a readable Markdown summary.
High-intent search queries and answers about installing and using this skill.
To automatically extract and caption YouTube frames, you need yt-dlp and ffmpeg installed. The process downloads the video, extracts frames at configurable intervals, and uses a vision language model to generate a searchable JSON index and Markdown summary of the video content.
Generating searchable metadata for long-form YouTube videos involves indexing each extracted frame with a vision model. This creates a structured JSON index and a readable Markdown summary, enabling frame-by-frame narration and scene understanding for extensive video archives.
Yes, you need to install ffmpeg and yt-dlp to download YouTube videos and extract frames. These dependencies are required to process the video file locally before applying a vision model for captioning and indexing the extracted frames.
Yes, you can use LM Studio for vision model captioning of video frames. The system supports LM Studio or Claude API backends to analyze extracted frames and generate descriptive metadata for scene understanding and frame-by-frame narration.
The frame extraction interval determines the frequency of captured frames, directly impacting the granularity of video indexing and scene summarization. A shorter interval captures more frames for detailed analysis, while a longer interval reduces processing time for long-form videos.
joegoldinDownload videos and analyze frames and transcripts with yt-dlp and ffmpeg.
nexu-ioGenerate hyperframes sequences with duration metadata for Remotion-compatible output.
ahostbrDownload YouTube videos, extract sampled frames, and generate VLM captions with timestamps.
blink-newAnalyze public video URLs and generate reports on visual and auditory content.
