normally that would be handled by using nodered as middle man.
There is an example on how send an image from robot → nodered → google vision api
here examples-files/object-detection at master · open-rpa/examples-files · GitHub
you will need to change the api from object detection to text detection
But you could also update the workflow to talk to abby, azure, ibm etc.
( and maybe here examples-files/ocr-with-google-vision at master · open-rpa/examples-files · GitHub )