* if an image doesn’t have alt text
* you need to be read the page
* you need to be described what’s happening in a video
A model built into the OS or browser seems like a no-brainer.
It does seem possible that this could be replaced with a local model in the near future. It's not clear the average user has the hardware specs for this to be an option today, but it will increasingly be plausible.
Keep in mind, though, that alt text is just one small part of making a web site accessible.
Siri does something like this when reading messages into your AirPods. It will give brief descriptions of photos sent in the message. I'm pretty sure it's all run locally.
How about localizing it into all languages supported by a major OS or browser?