Text this: Scene extraction accuracy of different models.