I would be careful with using the caption text, as the purpose for these two pieces of information are often going to be very different. More importantly, screen readers will already have the caption text available, meaning the potential for reading the text twice is high. More importantly, screen readers will already have the caption text available, meaning the potential for reading the text twice is high.
At the very least, if you’re going to do that, you should make use of the aria-hidden attribute, to keep screen readers from reading it twice:
<figure>
<img ... alt="This was the caption originally." />
<figcaption aria-hidden="true">This was the caption originally.</figcaption>
</figure>
But, if you’re going to that length then you’ll need to do very similar to some of the techniques described in the linked post below, and if you’re going to do that, you might as well use best practices.
For the alt text, the rule of thumb is to think of the purpose for having the image, and describe that purpose, or at times to just describe the image or the data within it. To illustrate the differences between a caption and alt text, with the former we might introduce the kind of data a graph is showing, and with alt text we might write out the numerical values of key pieces of information that the graph is illustrating visually.
The matter of writing good alt text aside, here is a series of ideas for how one could set up a system for adding alt text to images. One can be quite “lazy” as you put, just filling out a few metadata fields in the inspector and giving the image’s name in a binder item—but I would say even the styles are pretty easy to work with.
Of course for those using Pandoc → ePub, there is no need to set anything up as Pandoc does this on its own:
{alt="Here is the alt text."}
“image_name” is where you would link to the image itself somewhere else in the binder, to have Scrivener fetch it and insert it into the ebook, changing the hyperlink text to the image filename a necessary.
So in a sense it’s not too far away from the style-based approach, of giving this information on a line, we just use brackets around the different kinds of information instead of formatting styles.
(And if you leave the alt text off, Pandoc will use the caption text for the alt text, but it will also set aria-hidden, so this is okay, even though the image could probably be described better.)