Does anyone else have vaults of footage they want to search with an LLM, or is a producer scrubbing footage? I've created something so you can type "person in a red raincoat" or "sunset over water" and instantly get the exact frames you are looking for. It's also a secure offline environment, no uploading to meta, google or others.
I've just installed some tooling and python that gets this going:
Frame extraction with Python & FFmpeg Visionlanguage with Python and OpenAI's CLIP Embedding & lookup using Python and FAISS Interface with Python & Streamlit front end
You can get raw video folders to a searchable database you query in plain English. I'm about 90% there, and I'd love collaborators to help polish the UX and add functionality, I still get errors on some encoding types and random errors on some videos not extracting properly.
I think this is something many videographers and content producers could use.
I can share the install commands and Python scripts if anyone wants to help, or i can spin this up in GIT? Drop a comment or DM.