A dockerized tool to search the visual contents of a youtube video by text, using OpenAI's CLIP text/image embeddings. The results are then stored in a Postgres db in a vector format. You can add multiple videos and search through all of them with a single query.
approximately does the following things
youtubedl :: YouTube Video URL -> Video
decompose :: Video -> List[Video frame]
watch :: List[Video frame] -> List[CLIP embedding]
add memory :: PostgresDB (CLIP embedding) -> CLIP embedding -> PostgresDB (CLIP Embedding)
sentiment :: String -> CLIP embedding
recollect :: PostgresDB (CLIP Embedding) -> String -> List[(Youtube Video URL, timestamp, Video frame)]
There isn't really a UI at the moment so just visit the urls in the code to perform the actions.