Jagster reported a bug where their forum (using iPad, DiscourseHub, and OpenAI/GPT-4o) intermittently forgets automatic AI captioning despite having consent enabled. After testing various steps — including logging out, rechecking permissions, and uploading images — captions worked temporarily but failed again after restarting the app. Jagster noted this issue has persisted since the feature’s launch and speculated it might be device-, app-, or setup-specific, though few others seem to have reported it. keegan requested details on image properties (alt text, size, format), and Jagster clarified that 0.4MP is an outdated metric and suggested using pixel dimensions instead, also noting alt text limits may interfere with AI helper functionality. keegan deployed a fix (commit), and Jagster confirmed it now works — including on their own forum — though noted a minor formatting quirk where captions begin with ![<img src="[IMAGE URL]"> Kuvass…, possibly due to the OpenAI model. No further action deemed necessary unless the issue recurs or worsens.
My forum forgets that there is such thing as automatic captioning. It has consent to do it, but it acts like it just doesn’t know it and at some point it just doesn’t care.
That I did at the point where I didn’t get automatic captions any more:
I checked there is consent to do automatic caption; there is.
I logged out, closed DiscourseHub and opening it again and logging in.
I re-checked consent and it was there
I started new topic with image
I was asked permission to do automatic caption; it already has that permission though
Caption was created. I did two more testings in public and private categories and auto-captioning worked
I closed DiscourseHub, but not logged out. Then I opened the Hub and my forum
I started new topic with an image. I didn’t get a caption
I checked my settings and there was still permission to do captions
This isn’t new thing, but was there from very beginning when we get this feature. And no, I haven’t tested here or in playground-discourse.
What do you think — is this an issue of Discourse, DiscourseHub, iPad or my setup (it is quite basic without any fancy customizations). But because no one hasn’t complaining it means something is broken on my end, or… quite few uses or cares about automatic captions
BTW - 0,4 MP is not valid scale even dev-side loves it because it is easy to count. But it is still just markering term from early ages from digital cameras. You should use pixel sizes. That is more or less similar thing than using DPI.
Another thing to consider. ALT limit breaks that functionality when used AI helper, if I’m not totally wrong.
