Until AI models on the device will become a standard I really doubt you'll find a viable alternative, since then it will always be someone's server you send your pictures to, it's the matter of who you trust more, a well established corporation which will profit from your content or some shady company/startup that will do the same but in less controlled environment
I highly doubt if they really live stream the video you took, or pictures. I would much rather believe an OCR is being done locally and send it to a server for translation.