video2flow 0.1.0

Extract video frames as descriptive text flows for LLM consumption

video2flow

Extract video frames as descriptive text flows for LLM consumption.

Claude, GPT-4o, Gemini and other multimodal models can see images but not video. video2flow bridges the gap: extract frames → generate timestamped descriptions → feed the text flow to any LLM.

Installation

pip install video2flow

Usage

# Extract frames from a video
video2flow extract video.mp4 -o frames/ --fps 1

# Quick description (without vision API)
video2flow describe video.mp4 --max-frames 10

# Full pipeline
video2flow pipeline video.mp4 -o video_flow/

LLM Integration

Pass the output JSON to any LLM:

import json
flow = json.loads(open("video_flow/flow.json").read())
prompt = flow["usage"]["example_prompt"]
# Then: response = llm.invoke(prompt)

For detailed vision understanding, send image files directly to a multimodal model alongside the flow transcript.

License

MIT