My particular use case was that frames were generated on the fly and in great number so I wanted to stream them directly into a video instead of storing them all. Yes I realize this is a simplified version of the Tensorflow Streaming example, but I'd argue that it's good to have it in the "assemble video from sequence" section for clarity (e.g. it didn't occur to me to look at the TF case - I looked at the "assemble" section and found nothing helpful so I gave up on this article and looked elsewhere).