🔮 Sometimes I wish Mimi had Ice Cubes AI-powered Alt Text suggestions. Time saver.
@atog Yes. I’m just sad I don’t get to use it that much, since all my photos show up on Mastodon via M.b. Don’t really use the apps to post, only interact.
@maique Always wondered which app would do that first. Glad to hear that Ice Cubes beat everyone to it. Hopefully, that motivates others instead of some other random use of AI that they think is cool.
@pratik I tried it yesterday, after I saw someone posting about it. I’ve seen others raving about it as well. It’s really good.
@maique That would be sooo great! Not sure If it’s economically feasible though. I mean, I would subscribe for that (within means), but maybe not so many would(?) It could even save it as image an image description in Photos @samgrover
@maique @pratik @odd @atog I have a feeling that the use of AI-powered Alt text suggestions won’t end well. A lot of web pages and blog posts include barely relevant images that are there merely to break up the text and/or add a bit of optical variety. Mea culpa, I’ve been guilty of including such images myself, particularly when I used to post on Medium. Sometimes, as with @maique’s posts, you want to give a screen-reader user a detailed and evocative description of a carefully chosen, unique picture. In many cases, though, that detailed, evocative description would be extremely annoying or distracting: what you really want is to assure the the person relying on the screen-reader (or the person whose low-bandwidth connection prevents images from loading) that she’s not missing anything relevant or significant. Or so it seems to me.
@artkavanagh From the example @maique showed, I think it got the most parts right, even things I wouldn’t have thought about adding myself. I don’t see these texts as a complete solution, only a template for your own final text.
@odd The description seems very accurate, but
a) is that exactly what we want from Alt text? (It’s not how I’ve generally understood the concept but maybe I’ve been getting it wrong all these years), and
b) it didn’t give me any idea what the actual picture was like (though, because of my aphantasia, I wouldn’t expect it to).
@artkavanagh I’m reluctant to use it at the best of times. Like you, I don’t fully trust it - if that’s the right description.
@maique No? I forgot. It’s a one time purchase? I’ve got it, but went for the official one. I’m almost all defaults-defaults.
@r12t I’m not fluent in “code”, I’m afraid. I might look, but it would be as good as looking at any other language I don’t speak. Probably even worse, since I wouldn’t be able to get it into a translator app 🙂
@odd I’m still very much Mona, sometimes Ivory (almost zero these days), but love to use Ice Cubes once in a while.
@pcora Mona is nice. You can get it after the year is up. It’s not subscription on top of everything else.
@maique It works really well! I have imagined doing something similar on Mimi, for sure but yea, it would involve paying per use. Looks like Ice Cubes is using OpenAI's gpt-4-vision-preview and paying for it. So yea, I would have to alter Mimi's business model for any third-party API that does this. Alternatively, I could use a similar feature the Apple includes on device now or in the future. That's more privacy-aware as well. Another option would be for @manton to provide that as a service in the same way that he's doing for podcast transcription with Whisper.
@odd Mimi can't save data to your photo library because Mimi can't access your photo library. It uses Apple provided UI for image picking.
@samgrover But it is possible to add a…I don’t know what it is called…to ask for permission via the Apple provided framework? To birds with one stone as I see it. If I enter a really good image description for an upload, it would be great if my picture would get that in Photos as well (unless it already has a description).
@odd Asking for permission changes the privacy equation. Many people don't like to give access to their photo library like that to random apps (because it was abused by many apps, which is Apple provided their own UI). I'm one of those people :)
@samgrover I’m very wary of this myself, but I would trust Mimi/you with it. Not everyone has the same knowledge though, I can see how they might see a red flag, but I believe you could say yes to one and not to the other.
@odd @samgrover Hasn’t Apple now changed how apps access photos? It shows you the entire library but the apps have access to only photos that people import/add to the app.
@artkavanagh @odd I think you may be thinking of alt text as caption. I was making the same mistake for years. The sole purpose of alt text, I think, is to describe the photo in as much detail as possible. The example I posted did a great job and I don’t think I would have added that much detail either due to lack of imagination or pure laziness.
@pratik There are three ways. Apple still allows the old way (full access) and also limited library access way. But Mimi uses the way that has the picker that shows you everything in your library for you to pick media, but which gives zero library access to the app. It only provides the photo files you pick. No knowledge of other photos or photo library structure.
@samgrover Yup, that’s what I thought. But say, you enter caption in Photos you select to upload to Mimi, can’t it carry over that as alt text? I know the two aren’t the same but can let people opt in
@samgrover Any of the three would be cool, but of course I’d rather go for the last two. You’re right, he is using gpt 4, and paying for our use.
@samgrover About the GPT thing, I know of some apps where you can plug your own OpenAI token to use the AI parts and not incur a cost on the App developers. I think ShortCircuit does that.
@samgrover Yes! Maybe I’m super lucky and it will be approved by the time I wake up. Can’t wait to try it!
@samgrover Oohhh 👀. I just stumbled across the update to this thread. Any chance of posting the TestFlight link once it is up? I don’t know if I have it there anymore.
