A sophisticated local pipeline for video comprehension using ffmpeg, FunASR, and Qwen3-VL to analyze audio and visual content without cloud dependencies.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install local-video-understanding
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install local-video-understanding using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Local Video Understanding?
The Local Video Understanding skill is a specialized tool within the Openclaw Skills ecosystem designed for deep analysis of video files directly on your local machine. By leveraging a combination of industry-standard tools and state-of-the-art AI models, it allows agents to perform comprehensive content audits, accessibility checks, and scene descriptions.
This skill is particularly powerful for developers who prioritize privacy or require high-performance processing without recurring API costs. It bridges the gap between raw video data and actionable intelligence by integrating audio transcription via FunASR and visual reasoning through the Qwen3-VL model running on Ollama.
Local Video Understanding Use Cases
- Automated transcription of video content for accessibility and searchability.
- Generating detailed scene-by-scene descriptions and visual summaries of long-form video.
- Extracting key insights from local video archives for research or content creation.
- Building localized video processing pipelines that do not rely on external cloud APIs.
How Local Video Understanding Works
- Audio Extraction: The skill uses ffmpeg to isolate the audio track from the video file into a standard WAV format.
- Key Frame Extraction: ffmpeg captures frames at a specified interval (e.g., 1 frame every 10 seconds) to create a visual storyboard.
- Speech Recognition: The FunASR engine processes the extracted audio to generate a highly accurate text transcript, optimized for Chinese and English speech.
- Visual Reasoning: Each key frame is passed to the Qwen3-VL model via Ollama for detailed image-to-text descriptions.
- Intelligence Synthesis: The system combines the transcript and frame descriptions to provide a holistic understanding of the video content.
Local Video Understanding Setup
Ensure you have ffmpeg installed in your PATH and Ollama running with the Qwen3-VL model.
Environment Setup
# Create and activate the FunASR environment
conda create -n asr-local python=3.9
conda activate asr-local
pip install funasr modelscope
# Pull the visual-language model
ollama pull qwen3-vl:8b
Directory Configuration
Ensure a cache directory exists for model storage, typically at C:/Users/TOM/.cache/modelscope for Windows users.
Local Video Understanding Data Schema & Taxonomy
The skill manages its output through a structured file hierarchy and specific metadata points:
| Data Type | Format | Storage Location |
|---|---|---|
| Audio Stream | .wav |
Root project directory |
| Visual Samples | .jpg |
/frames directory |
| Transcripts | String/JSON | Captured via FunASR output |
| Scene Descriptions | String | Generated by Qwen3-VL via Ollama |
All temporary assets are organized to allow Cloud LLMs to ingest the final text-based summary easily.
Local Video Understanding Advanced Features
- Multi-model support for specialized speech recognition using FunASR over standard Whisper models.
- Customizable frame extraction rates to balance between detail and processing speed.
- Support for existing subtitle file detection (.srt, .vtt) to bypass ASR when data is already available.
- Seamless integration with Cloud LLM APIs for high-level thematic analysis based on local extraction.
SKILL.md
Loading
Related Openclaw Skills
Boxed HTTP Server
A lightweight, WebAssembly-sandboxed static HTTP server featuring built-in basic authentication and reverse proxy capabilities.
Workspace Browser 3.0
A high-performance workspace exploration tool featuring syntax highlighting, global search, and integrated AI code explanation persistence.
Anime Social Portrait
An advanced AI skill that converts real-life photos into stylized anime-style avatars, covers, and headers for social media profiles.
Text to Free Video - Generate Videos from Text
A powerful AI video generation skill that converts text prompts and documents into professional 1080p MP4 clips using cloud GPU processing.
Deep Content Search
A high-performance search and extraction tool for aggregating content across 8 major Chinese social and media platforms.
Gillette Encyclopedia Skill
A comprehensive knowledge module providing deep insights into Gillette brand history, product lines, and global market positioning.
Featured*
