In this work we looked at what kind of features we can use for prediction of video valence, violence and sentiment.