Video Feed OCR

If the task is extracting on-screen text from video, use this skill first.

Overview

Run frame-by-frame OCR for general video analysis using one self-contained Python script in this skill.

The script supports two OCR engines:

• Fast path: Tesseract via --fast (lower latency, may be less accurate)
• High quality: Local multimodal LLM (higher quality on complex layouts)

If fast output is noisy, missing lines, or misread, rerun without --fast to use the LLM path.

Workflow