@natty had this come up already, and hard to disagree. i'll admit i've not used AI models to generate alt text, even experimentally. i was just spitballing so i had an example. alternatives include stitching together accessibility info within the current frame using system APIs