Video Analyzer

Analyze video files by extracting keyframes with ffmpeg and using vision to understand content. Supports single and batch processing.

Analyze video content using ffmpeg frame extraction + vision AI. Supports single files, batch processing, and structured reports.

Prerequisites

• ffmpeg installed (pkg install ffmpeg in Termux, or apt install ffmpeg)

Scripts

extract_frames.sh

Single video frame extraction.

bash

bash scripts/extract_frames.sh <video_path> <output_dir>

Outputs frame file paths (one per line). Density adapts to duration:

• < 30 s → 1 frame/s, max 30
• 30 s – 5 min → 1 frame / 5 s, max 60

• 5 min → 1 frame / 15 s, max 40

analyze_batch.sh

Batch video processing with metadata extraction.

bash

bash scripts/analyze_batch.sh <video_file_or_directory> [output_dir]

For each video, outputs:

• metadata.txt — duration, resolution, codec, size
• frame_XXX.jpg — extracted keyframes

Workflow

1. Extract — Run the appropriate script
2. Analyze — Pick 3–5 evenly spaced frames, pass to image tool with prompt: "Describe this video's content based on these keyframes"
3. Report — Summarize findings in structured format

Report Template

markdown

## Video Analysis Report

**File:** <filename>
**Duration:** <duration> | **Resolution:** <WxH> | **Size:** <size>

### Content Summary
<concise description of what happens in the video>

### Key Observations
- <observation 1>
- <observation 2>
- <observation 3>

Tips

• Clean up frames after analysis: rm -rf <output_dir>
• Common Android video locations: /sdcard/Movies/, /sdcard/DCIM/Camera/, /sdcard/Pictures/WeiXin/, /sdcard/Pictures/QQ/
• For audio transcription: ffmpeg -i <video> -vn -acodec pcm_s16le -ar 16000 output.wav