Poor results for Multi-subject Personalization

Hello, please could you share some pointers on how to use MM-diff for multi subject personalization? Wanted to cite this work in our benchmark, but I am getting very poor results for multi-subject. is there any way to have varying trigger token names for different people

The issue names MM-Diff and multi-subject personalization but no files, tests, or entry points. Start by reviewing the repository's usage guidance and personalization flow, then reproduce the reported multi-subject results and inspect how trigger token names are handled. Done means the expected usage is documented or the requested varying-token capability has a defined, tested implementation scope.

Written by the indexing model from the issue text.