Interesting! Certainly looks like a bug.
Can you try prompt engineering around it? Just tell the persona only to use English when describing images?
I will look at a more comprehensive fix.
3개의 좋아요