Grigorij-Dudnik/video-understanding-local — reverse-engineered prompt

Reverse engineered prompt

GitHub

Build me a Python tool that can analyze videos completely on my own computer so my footage stays private. I want to give it a video file and get back a useful written summary of what happens, using both the spoken audio and what is visible in the scenes.

It should have a simple function I can call, like analyze_video, with an optional custom instruction so I can ask things like “list the tools used” or “make a short filename for this clip”. Please make it handle longer videos by breaking them into chunks, transcribe speech, inspect frames, and combine everything into one clear answer.

Also include an example script that goes through a folder of raw videos and renames them with descriptive lowercase names based on their content. Make the setup straightforward with a requirements file and README, and note that it needs Python, FFmpeg, and a CUDA capable GPU. Look up current model docs online if needed.

Want more depth? Deep Reverse