API to run task over series of (response, image) pairs

[geoffroy-noel-ddh]

e.g. execute rotation of the image based on answer by model to question about orientation.

or crop region, split double pages, etc.