Sounds like OP wants an LLM-based solution, which is not something built into BookStack by default.

Danswer might be an option here. It has a confluence connector (and I also wrote a BookStack connector for it). If have a video of me using it (with BookStack) here at about 7:55. When I used it last it only worked with OpenAI but I think it supports local models now.

No worries, although the color of them is now considerably blown out and patchy due to the compression, and they're still relatively large. It's the high pixel size that's really affecting things. If you took the originals and resized them to 1168px width (the display width on that page) at about 75% quality (still JPEG) you'd probably end up with much better quality than you have now, and at lower file sizes. Squoosh is pretty good for playing around and comparing results.

Congrats on the marriage!

BTW, those wedding photos total about 31MB, pretty massive for a webpage. Might really lag for folks on slower hardware and/or networking.

Happy to help! Just as an extra warning, don't auto-upgrade BookStack containers via something like watchtower. Most of the annoying issues I've had reported from docker users of BookStack seems to be due to auto-upgrade failing, which can often be hard to track down since errors may not show right away, or the user many not realise an update was attempted.

This is common on windows. Something funk goes on at the filesystem level in this configuration which causes issues with migrations. There's a relevant thread here. Some folks in there have solved it via alternative volume options, or alternative database docker images. Should not be something you experience on Linux though.