Can preference date be used in other text-to-motion models?

This is truly an innovative piece of work!
Regarding the preference data, I would like to ask: if I want to use this data to optimize the generation results of other text-to-motion models, such as diffusion models, would that be feasible?