I do a lot of this work[3] (web to documents) and it's interesting to see other approaches. The medium image problem is something I've faced as well, but never got around to fixing. I'm planning to get a Remarkable soon, so will definitely be trying this out.

My personal solution has been https://github.com/captn3m0/url-to-epub/ (Node/readability), which I've tested against the entirety of Tor's original fiction collection[0] where it performs well enough (I'm biased). Another tool that does this beautifully well is percollate[1], but it doesn't give enough control of the metadata to the user - something I really care about.

I've also started to use rdrview[2], which is a C-port of the current Firefox implementation of "reader view". It is very unix-y, so it is easy to pipe content to it (I usually run it through tidy first). Quite helpful in building web-archiving or web-to-pdf or web-to-kindle pipelines easily.

[0]: https://www.tor.com/category/all-fiction/original-fiction/

[1]: https://github.com/danburzo/percollate

[2]: https://github.com/eafer/rdrview

[3]: https://captnemo.in/ebooks/