All sounds pretty sensible. I do think it might feel annoying waiting minutes to download a model for the sake of generating a line of alt text the first time, though. It would probably be quicker to write the alt text.

I think if you read through this and take it at face value, there is a pretty clear picture of what happened: https://robmensching.com/blog/posts/2024/03/30/a-microcosm-of-the-interactions-in-open-source-projects/