Given a sample of text, an LSTM layer learns the style and starts generating the text (character by character) in the style of the author. The output, given a random seed is as shown when trained on a passage of Nietzche:
oral to be in the his art in the sublict and in the subler and and to the ore and the his to the his been the for the his pore of the some the some and and the the the has the respending the some the here to the here the inderstand and to the some the some the some and the presance and the some the his an and the some his to the himself the are are the himself to the some his to the here to the lo