Autorecognize text in image for Alt-Text

Although Discourse has a slick feature to edit alternative text image descriptions in the preview pane (for those who are wary of Markdown editing), in our community of 1800+ almost no one writes alt-text. The file name is pretty much useless in my book…

I know there is hope/possibility for AI to do a first cut, and I can see that as some minimum bar, I prefer human authored alt text for other low vision humans to read. It does not “solve” the problem, but wonder about the ability I see in Mastodon to have an option to OCR the text from an image and insert in the editing pane.

Here I have uploaded an image in Mastodon, that has text, and I am preparing to write my Alt text.

When I click Detect text from picture it only takes seconds to populate the text entry field:

Minor editing might be needed but it save a lot of time. And it’s better then a screen reader comeing across this and speaking “Screen Shot 59.jpg”



Referenced this feature request on the OpenAI Discourse forum.


We’re working on a feature that will use AI to generate text descriptions for images, so we’re jumping over the OCR step for now. We’ll announce it on Meta once it’s available!