Alt text is the field everyone knows they should fill in and almost nobody fills in, because writing two thousand descriptions by hand is not a task anybody starts. So it stays empty, the library stays unsearchable, and the accessibility audit stays uncomfortable.
The Metadata Assistant writes a first pass for you. Titles, alt text, captions, descriptions and tags, in the language you choose, across as many images as you select. What was an impossible job becomes an afternoon of correcting.
You choose the engine. A local model runs on your own machine with no key and no cost, which is the right default for a big library and for anything involving people. OpenAI and Google Gemini are available if you have a key and want their phrasing.
One honest caveat, and it is the important one: a generated description knows what is in the picture but not why the picture is on the page. It can see a person at a desk; it cannot know that the photograph is there to show the desk. So treat the output as a draft, and correct the images that carry real information. That is still an enormous saving over starting from nothing.
