Abstract:Complex language models trained on huge text corpora have shown unparalleled text generation capabilities, and thanks to transfer learning, are accessible to a greater number. However, despite recent developments, users are not yet able to fully control particular aspects of the text produced. This is why we propose a finetuned OpenAI GPT-2 model for controllable and contextualised text generation specific to novels. By integrating it into a web-service, we would like to enable authors to write and ask for automatic text generation which is consistent with both previous and next paragraphs. They can specify the genre of their book, the length of the desired text, the entities it should mention and its content via keywords or a short summary. We explore the technical possibilities and limitations around these objectives.