Cartoons are a special snowflake: Jeremy Howard said in the fast.ai course that you probably need to retrain multiple layers of convolutional network to handle them. I.e you have a different model for them than for photos.