I believe it's the nightly build on windows currently. Actually just looked and came back I'm mistaken it's on mac looks like too. Not sure why it's not working.

Yeah, that's the irony here.

Article: "Listen to this overcomplicated warranty registration process with some jabs at overcomplicated IoT bullshit. They print the custom unit serial number on the sticker anyways, why not just also customize the QR code to also embed the serial number?

"Anyways, here's an overcomplicated way that uses AI ML to parse that custom unit serial number from a picture.

"Oops, it doesn't work for complicated browser jerry-rigging reasons."

chrome://flags/#prompt-api-for-gemini-nano to enable this on current Chrome (at least on Linux). It worked after I enabled that flag and restarted Chrome, though it required two tries (apparently it needed to download a model).

I tried to figure out if this stuff was available in other browsers but unfortunately came up short.

Googler, opinions my own.

The Prompt API [0] works in Edge, though it uses the Phi-4-mini model instead of Gemini Nano. [1]

Currently it doesn't work in Brave (at least on my machine), and I can't find anything online suggesting whether they plan on supporting the Prompt API. You can go to brave://flags/ and it shows "Prompt API for Gemini Nano" and "Prompt API for Gemini Nano with Multimodal Input", but it doesn't seem to actually work.

0. https://chromestatus.com/feature/5134603979063296

1. https://learn.microsoft.com/en-us/microsoft-edge/web-platfor...

PromptAPI seems to be the usual fire-and-motion Chrome move.

It doesn't work for me in Chrome either. Oh irony.

Same on Vivaldi, macOS.

Same on Fennec

[Author here] Safari is not supported. See https://developer.chrome.com/docs/ai/prompt-api#hardware-req... for the requirements.