I added more of the Cgi model to the dataset. 0.8 strength is what I believe is best but 1 strength does still work.