[mrkn1]

I made something very similar 2 weeks ago, re the upcoming OpenAI phone.

https://news.ycombinator.com/item?id=48040327

[ZeWaka]

The image processing is neat. Local model ran in the browser?

[mrkn1]

thank you! actually it's an API call to a VL model on Deepinfra (model is Qwen3-VL-30B-A3B-Instruct)

[danielrmay]

This is really neat, and disturbing.

[mrkn1]

thank you and alas yes, the image understand is the only LLM, the rest has been available on browsers through js since the 2000s