The double-cmd-C to trigger it is a nice touch, no separate hotkey to remember. What sold me is that it runs the model on-device, no account or API key needed for the basic case, so I didn't have to think twice about pasting anything sensitive into it. The screen-region OCR mode for text you can't select (PDFs, images) worked well in my quick tests too.
Thanks Gal. Good catch on the drag-a-box mode - right now the popup only shows up once the text has been read, so the ocr pass looks like pause. A visible "reading the screen" state before translation starts is a small change and its already on my list forupdate.
Out of curiosity, which Mac you were on? OCR step is noticeably quicker on M2 and later, and I'd like to know how the M1 experience feels for people other than me.
And thx for the DeepL comparsion, the alt-tab habbit was exactly what I built this to get rid of.