Comparing AI-Powered Accessibility Tools: PiccyBot, Be My AI, and Aira's Access AI

This one is still in beta, but in a lot of ways, it's already turning into my go to after Piccy bot. I was able to use it this morning to share the description I was given of a photo my pastor at church shared on facebook. I know I'm super slow, but I just figured out how I can save FB pictures I'm interested in to my photo library and then use Envision AI, Piccy Bot, or whatever to describe them. Wish that the process was a bit more streamlined, but I'm thankful for the ability none-the-less.
Getting back to Envision AI, the ability to more or less fully customize the assistant is a huge plus for me. There are numerous voice options to choose from, and it basicly lets you write what you want the personality to be like, although I'm not quite sure I've done that correctly yet. Yes, some customization is available in piccy bot as well, but it's far more limited in that one area. At any rate, I have all of these vision assistants in my virtual toolbox, and it's sometimes fun to compare the info each one provides.

Envision Assistant Beta, is My go to AI assistant, One can take a image have it described just like, Be My Ai, with the added benefit of asking questions with voice, the AI voice of choice replys

seemed like a broken thing till a few days back but since couple of recent updates, is shaping up to be an incredible thing. It's exactly what it claims to be--an all-in-one AI assistant developed specifically for the visually impaired.
Interestingly, since you can give a description of your requirements during account creation, it acts with a lot of context awareness.

What do you mean by, you're able to describe your own style?

During profile creation, you can give a description about yourself and what kind of personality you like for your assistant. For example, in my case, I told it that job-wise, I'm a civil servant and in that context, I need factual, matter-of-fact info while out of it, I'd like my assistant to be humourous and a bit sarcastic. And exactly that's what it's doing. Also, since I told it that my jurisdiction of work is Tamil Nadu, India, it is framing all info with that context in 'mind' (don't know what other word to use).

What happens when you ask your friendly neighborhood AI to describe images with:
1. An orange
2. A lime
3. A violet
4. An orchid
5. A rose
6. A peach
7. A cardinal

Heh, I could go on and on, but yeah, all items listed above are both a color, and a thing. 😇

Since the AI is intellegent enough to describe a picture in detaile, it should be intellegent enough to distinguish between a colour and a thing. So nothing unexpected happens I guess.

@Assistive Intelligence said, "With a personal prompt or style, you could tell it to not use the language of colour in its responses."

My question was a direct response to that. I was hoping for some honest feedback here. . .

@Gokul, Appreciate your response. Here is to hoping you are correct. 🙂

I am correct that is. Tested it with gpt 4o. Told it I'm blind and that it shouldn't talk to me in the language of colour. Then attached the picture of a rose and asked to tell me all about the picture. It came up with "This is a rose flower" bla bla bla... Guess this was your concern. Will be interested in testing this by recreating any scenario that anyone can suggest as best as I can.
That's why these things are called large language models. They are capable of understanding dynamic use of language. That's why they're able to respond to natural language queries. Therefore, if at all there's a problem of the sort suggested with these models, it should be treated as a fundamental bug and needs to be reported to the company concerned. (I'm primarily talking about models like gpt and claude, not the stupider ones like Meta AI).

a unique way to look at it for someone who had sight at a point. @AI I suppose you should try it. I mean, makes sense as to how it'd change the way visual info is conveyed...

Just to make it more pointless and twisted, you could have the AI describe people in terms of if two famous people from the mid-80s or earlier had children together...
I sometimes have to look up the release date of a movie just to make sure whether I watched it with eyes, or what ever it's called that my brain does now when watching a movie.

Other people had mustaches in that TV series though.

I think Rick had one for a while, but Higgins and TC did all the time. And TC was the name of the helicopter guy.

hi Family. i would be interested to know details of the extent of ios shortcuts support of the ai apps.
which one of them allows sending files or photos and getting text back into the shortcut for further processing

than you! awesome discussion. especially the magnum part - harhar

I've only used be My AI so far. Works well with my air conditioner remote. I observed be my AI did a very good job of identifying the remote control, including the brand name. Does better reading the LCD or LED display than regular Seeing AI or InVision.

Located all the controls and told me where they were relative to each other.

Was able to tell me when night mode is turned on which is indicated by a crescent moon on the display.

I use the app every day to set my air conditioner. I asked if it would remember this remote and its layout for the future but be My AI does not do that. I would like be my AI to learn the devices I use, and how I use them for the future.
Sometimes I have to remind it that the night mode indicator is on the display because it sees the button has the same symbol. But the button doesn't Light up.

I would really like these assistance to remember my stuff and how they work.

It would also be nice if Be My A I was voice controlled and more interactive. It's a bit of a problem taking extra pictures and entering text and submitting it with voiceover.

But it's a really good start and I'm excited to see how well these tools develop.

.

I've these 3 apps assigned to Action Button, and bottom left and right on lock screen (where flash light and camera use to be). It's great that we can customize the lock screen buttons and Action button to get access to visual info in different ways. Depending on context, I can choose if I want to quickly use PiccyBot, AIRA Access AI, or Be my AI and not have to unlock my iPhone.

I think Claude for iPhone needs to be added to this article, but is there anything else on iOS?

I wish the image description feature would come to Voiceover at some point. This feature is a TalkBack feature and the image processing is done by Google AI. You get a detailed description of any photo that TalkBack is focused on. Once you are focused on an image, you bring up the TalkBack menu and tap describe image and the results are equally as good as Be My AI.

Upgrade it, learn to use it (again) and chekc this out. I guess I won't need a GeminI Plus subscription?

@Lotti no you don't. But it doesn't allow you to take a picture using the camera and get it described.

Not quite. But anyway...I've used most of these apps with mixed results. I've also used the native camera app. I for one, would greatly appreciate an in-depth audio walkthrough of all these that are available to us thus far: pros, cons, and just examples of how to get a good enough aim in order to snap a picture/image of say, ingredients on a box or can of something. Plus any and everything in between, a "photography for dummies" kind of thing. I for one am a bit baffled by all this. Just the fact alone that the built-in camera app is usable by a blind guy like me is still a bit daunting. I'm not trying to complain here by any means; I'm amazed and impressed that the built-in camera app is so accessible.

If I wote some things up and got Notebook LM to make a podcast, would you give it a chance? Have you heard any of the NLM output? They are quite remarkable.

I've not heard of them before, but I'd love something like this and would definitely give it a go.

I just tried this tool from Google. Although it is a bit tricky to use with a screen reader and many of the buttons are unlabeled, the results are amazing!

As a quick summary for those who don't know what NotebookLM is, it is an AI tool into which you load up information that you want it to process, ask it to summarize, and/or ask follow up questions. The difference between NotebookLM and many other AI tools is that NotebookLM just processes its responses from the input you give it. Thus it doesn't scour the entire internet or rely on its own trained model.

Besides the ability to input copied text, PDF documents, audio, and other content from Google Workspaces, one truly remarkable ability is that it will generate audio of a two person interactive dialog discussing your input. Truly amazing and a bit scary. The voices are superb and it sounds like a real podcast.

The tool is currently in development and free to use, but things could change.

--Pete

You had me at PDF. 💘

The more I read about it and the more examples of it I hear, the more significant I think it is. I know there are some great podcasts out there and I don't think this is going to replace those, but for us?

As Brian said, PDFs!

LLMs have been the hero that slayed the PDF monster for me, but NotebookLM might be the tool that turns PDF into my fav format!

I've put a 'odcast' about Glidance on my Mastodon feed, for anyone to try...I'll see what I can do with this article about AI, along with some of the others. It might take a few goes to get it right.