The only issue I see in these is that "tags" (in #3) is a freak-out word for some people, but that's a minor problem. I like all of these options, so I'm just going to call out the things I really, really loved: Empty annotations are a great idea, and the icon is amazing. Running OCR on images would also be super cool, and it might make Wikisource poke their heads up and wonder how we can help them. 7 and 8 seem like super-natural extensions of our existing workflows, and we definitely had thoughts about tying this in anyway.

There are many interesting patterns here to consider. One aspect I found it may be missing, is that I'm invited to add an annotation but I'm not told why I should do so. We probably don't have room for an in-detail explanation (which is probably not worth it in any case since these tend to be ignored), but we can connect the messages we use with the purpose a bit more. For example, "Would you like to add an annotation to better describe the image?", "Would you like to add an annotation to connect parts of the image with relevant information about them?", etc. The example 1, can use some of these concepts to avoid the guided tour to just replicate the main action label.

Another aspect to consider is how much we want to present this as a blocking step. I think it makes sense to present this as a convenient option for those interested but not add friction or get in the way. This is my main concern about the use of a modal in 7, although a non-blocking alternative would make sense. The rest seem to integrate fluently with existing processes, although we may want to consider options to discard the calls to action in case users want to focus on the image.