This functionality would be very useful to my work as well!
The rationale behind this is that ImageAnalysis seems to do the highest quality OCR, but it doesn't output positions. So in order to get the positions, I have to use VNRecognizeRequest, which seems to offer slightly lower quality OCR.