NV Access

3 weeks ago

NV Access
3 weeks ago

Hi everyone, an exciting announcement today - we've released our first 64-bit alpha build, which ALSO includes on-device AI image captioning. For the full details and the caveats - please see this post in the NVDA user group: groups.google.com/a/nvaccess.o…

(Also, in response to a couple of questions:
- The keystroke for image captioning is NVDA+Windows+comma
- And yes, you can setup the new alpha as a portable copy rather than installing it).

#NVDA #NVDAsr #AI #ImageDescription #Image #64bit

NVDA 64-bit with on-device image description alpha build now available

^{groups.google.com}

#nvdasr #AI #image #nvda #imagedescription #64bit

Tamas G reshared this.

in reply to NV Access

Bill Dengler

in reply to NV Access 3 weeks ago

If 64-bit has indeed hit the alpha channel, you will want to uninstall any NVDA you have on the machine. 32 and 64 install two different directories and different registry locations, and so you may have both on the machine at once if you don’t uninstall first. Your user config will persist.

This entry was edited (3 weeks ago)

in reply to NV Access

Alex Hall

in reply to NV Access 3 weeks ago

That's really neat! I don't know what you're using for image descriptions, but 230MB feels small. For those with a lot of room to spare, could there be an option for a larger model? I guess communicating the increased system resources needed to make use of such a model could be a UX challenge.

in reply to NV Access

Amir

in reply to NV Access 3 weeks ago

Great, but NVDA+Windows+,? This is not a good choice IMO, and something better like SHIFT+NVDA+G could have been used. TTS engines don't announce the comma in NVDA+Windows+, either.

in reply to Amir

NV Access

in reply to Amir 3 weeks ago

@amir I'm not sure why that keystroke was chosen, but even if it stays that way, you'll be able to change it - and yes apologies for writing the symbol rather than the word in the original email - hence was I explicitly wrote it in the post on here.

@Amir

in reply to NV Access

SirMars

in reply to NV Access 3 weeks ago

@morganlynn1992 Oh nice. Very exciting to see the progression of the 64 bit versions, and I am not worried. add-in's will get there eventually.

@Morgan Pimentel

in reply to SirMars

NV Access

in reply to SirMars 3 weeks ago

@SirMars @morganlynn1992 Re AI image description & add-ons, there are already a few - some have been available and in developement since at least 2019. So, if you don't like what we're trying out, you can definitely keep using the existing add-ons (well, noting that they won't currently work with 64-bit NVDA at this time, but that's just because of where we are in the cycle - those add-ons work with NVDA 2025.3 and will work with 2026.1 by the time it comes out)

@SirMars @Morgan Pimentel

in reply to NV Access

SirMars

in reply to NV Access 3 weeks ago

@morganlynn1992 Oh ya, I was just warning others. No worries. That is why it is on a test machine. and not production machine. And thank you all for your hard work. Can't wait to see the final project at some point.

@Morgan Pimentel

in reply to NV Access

Majid Hussain

in reply to NV Access 3 weeks ago

is that an llm model being used?
is it a custom model or something?
curious

in reply to Majid Hussain

NV Access

in reply to Majid Hussain 3 weeks ago

@mhussain Yes, The model we are currently shipping with is Xenova/vit-gpt2-image-captioning, however we plan to add the option to download more models. Newer and better models are always coming out, so we've designed the code to be easily upgraded to the latest and best models.

@Majid Hussain

⇧

NV Access 3 weeks ago • •

NV Access
3 weeks ago