Trained on a lot more images
Version 2 of my first ever model, trained on twice as many images so hopefully more reliable/consistent appearance
