Marcin Wichary4d ago

I’m curious: Did people/apps ever attempt to bundle alt text with the image as metadata? (Like EXIF?)

That feels like the right place to put it, so then it could be read by apps/browsers, and also travel with the image on save etc. Rather than specifying it as out-of-band text like we do today.

Show threadEthan Marcotte4d ago
@mwichary I think one challenge is that alt text is best when it’s contextually tuned; it’s not static. (The AFB did a great job explaining this, though it’s in the context of a bigger piece on LLMs: https://afb.org/blog/entry/alt-text-age-ai)
Beyond Alt Text: Rethinking Visual Description in the Age of AI | American Foundation for the Blind

The American Foundation for the Blind
Show threadMarcin Wichary3d ago
@beep Thanks! I worry that some of these arguments are letting perfect be enemy of good, though…
Show threadMarcin Wichary
@beep (I don’t mean the AI stuff, mostly just expecting alt texts to be immaculate.)
Apr 10 at 12:01amWeb
Show threadEthan Marcotte3d ago
@mwichary Oh, I don’t think there’s any such thing as immaculate alt text! I thought Eric did a good job outlining some of the potential drawbacks/harms, though: https://ericwbailey.website/published/thoughts-on-embedding-alternative-text-metadata-into-images/
Thoughts on embedding alternative text metadata into images

The idea of “solving” alternate text descriptions by automating them away so that they are not a consideration is a bad frame.

Show threadMarcin Wichary3d ago
@beep Thanks! I think the key difference is this maybe: “Not having the ability to adjust this baked-in content—or be educated that its quality is even a problem in the first place—means it is far more likely that the same low-quality content will be passed around.” In my head, the UI to adjust text exists as today, just the information (alt text) is more resilient.
Show threadMarcin Wichary3d ago
@beep In other words, this feels like a classic challenge that people think in very abstract systemic ways, but do not consider how user experience can factor in and improve the system or the usage.
Show threadMarcin Wichary3d ago
@beep I disagree with a lot of what this author wrote, but appreciate you sharing this!
Show threadScott Kellum3d ago

@beep @mwichary I see this as a “yes, and” kind of feature. I would love to see a “content cascade” where alt text can be in EXIF data, and then overwritten with alt text. Like with CSS, style attributes override the stylesheet; alt text can override EXIF data.

“The process of adding a new category of metadata to even a single image standard is a Herculean undertaking.”
— EXIF is a cross-file-type standard that exists, and adding one field seems very doable to me. Font specs regularly add to their naming tables; I don’t see why image specs can’t.

I see this as another tool to help with some workflows, not a change to how we do things. Adding alt text to EXIF data seems like a useful tool to have in addition to the wonderful tools we already have.

Show threadMarcin Wichary3d ago
@scott @beep Yes, all this is how I feel, too.
Show threadEthan Marcotte3d ago

@scott @mwichary One point I take away from Eric’s post is that the surface area for any given image is much, much more vast than any one medium: overriding EXIF data with alt text isn’t going to help someone who’s, say, sharing an image over text.

(I’m not even going to get into the very good points he raises about abuse and localization, which have basically haunted me since I first read the piece.)