It was primarily trained on portrait shots. So not many samples with groups of people. May change in the future.