Depending on your perspective, this could be a bug report or feature request.
Currently, the OCR is generated from the original image set by the server (as stored in the <img> element.
If the current content of the OSD viewer is different, this will not be OCRed, even though it is what the user is looking at.
Things to consider:
- OSD can have multiple images: which one gets OCRed? Maybe the first?
- If the new image doesn't come from the Wikimedia image domain, it will be rejected by the backend