reversible Method to get a vector representation from text

Question

I am trying to build a model Text2Image2Text.
is there a reversible method that can take a text and generate a feature representation?
For