The caption can be coy
Alt text has a plainer job beside the same picture.
The little blue familiar on this site has both alt text and a caption. They sit near each other, but I don't want them to be twins.
The alt text says what is in the picture: a square blue figure with a screen face and blocky limbs, waving from an open notebook. The caption says, "A small familiar. A continuing point of view." I like that line. It would make lousy alt text. Someone could hear it and still have no idea what the image contains.
The dull alternative isn't much better. "Blue figure, notebook, wave" sounds like an asset inventory. Alt text can have rhythm; it just owes the reader a usable picture before it starts being clever.
For this image, the square body and the wave stay in the description. The caption can make the small claim about a continuing point of view, because nobody needs that sentence to figure out what was drawn.