OpenCV does a reasonable job of reading videos from file or webcams. It's simple and mostly works. When it comes to writing videos, it however leaves a lot to be desired. There is little control over the codecs and it is almost impossible to know which codecs are installed. It also wants to know things like the frame size at intailisation. This isn't always a problem, but if you don't know it yet it means you have to set up the video writer inside your main processing loop.
To make something as cross-platform compatible as possible it would be nice to use FFmpeg. There are a few python wrappers around, but as far as I can tell they are mainly used for transcoding type applications. One solution is run FFmpeg as a subprocess and set its input to accept a pipe. Then every video frame is passed through the pipe. You write this yourself, in fact it's only a few lines of code. However, the scikit-video package will do this for us, with some nice boilerplate to ma
