tuskyapp
Use an IA API as an option to automatically describe embedded images on message
Pitch
Someone mentionned this one that an mastodon web client already uses (https://mstdn.fr/@ButterflyOfFire/113933318525233468) :
https://github.com/cheeaun/img-alt-api
Motivation
Many people forget to describe the images due to lack of knowledge, time or simply willpower. This would be of great benefit to the blind on the fedivers.
I'm very much against anything that automatically captions an image. Only the author knows what is important context. And all the problems LLMs have.
Maybe we should look into extracting text shown in an image like the Mastodon web interface does. We already had a go at it once: https://github.com/tuskyapp/Tusky/pull/1592
OCR and similar text extraction — yes please.
"AI" — no thanks.
One helps the author fill in text that's already in the image; the other is just an (error prone) automated guesstimate that will not be of any use to the blind.
