First, create a project for each of your writing topics.
You only need to simply set your writing intent (of course, you can also make detailed and complex settings).
General books and monographs
When creating a project, you need to choose whether to create a general book or a monograph.
The core differences between monographs and general books are reflected in two aspects:
- Chapter structure: Monographs have a relatively strict, fixed-structure chapter and section table of contents.
- Citation style. Monographs must include citations.
If you choose to create a monograph writing project, you need to further select a citation style option:
- Unified numbering for the entire book (all references in the book are placed at the end of the book);
- Independent numbering by chapter (references for each chapter are placed together at the end of the chapter);
- Independent numbering by page (references for each page are located at the bottom of the page. This feature is only available when exporting Word files)
Note:
Because the references used for AI writing in this project (such as EPUB format) are not paginated and have no concept of “page numbers”, while PDF files do have page numbers but OCR tools usually cannot extract them correctly, the citation annotation capability in this project is limited.
For details, please refer to the help document “Known Issues and Limitations”.
Using non-text elements
Whether you create a general book or a monograph, you can choose whether to allow AI to automatically generate and insert the following elements during the writing process based on context:
- Charts (SVG format);
- Formulas (LaTeX format);
- Mind maps (Mermaid format);
- Images (extracted from reference materials).
Writing goals
- Total word count: roughly how many words per book;
- Number of chapters: roughly how many chapters per book;
- Number of books to generate: how many books in total.
Note:
The total word count and number of chapters for each book are only expected targets; the actual generated book may be slightly more or slightly fewer than the values you enter. This depends on how many reference materials are provided and on the AI's independent judgment in topic planning and main text writing.
When the materials you provide are too few to support rich writing content (for example, you provide only 3 reference materials totaling 100,000 words but want to create 5 books totaling 1,000,000 words), the AI will determine and indicate that the materials are insufficient to support the writing goal, and will stop subsequent work.
Creative Freedom
Almost all large language models use sampling temperature (temperature) to control the model's creative ability.
- The lower the temperature → the more deterministic the output, the closer to the reference materials, and the more conservative;
- The higher the temperature → the more divergent and creative the output, with more supplements and associations.
Creative Freedom is essentially mapping a level selected by the user (0–4) to the large language model's sampling temperature (temperature), which is then passed to the AI to generate content in different writing stages.
In this project, the approximate characteristics of the 5 Creative Freedom levels are as follows:
- 0-Faithful to the original: closely follows the materials, quotes extensively, and barely improvises;
- 1-Conservative: focuses on the materials with a small amount of supplementary explanation;
- 2-Balanced (default/recommended): balances faithfulness and readability;
- 3-Creative: moderately expands and gives examples based on the materials;
- 4-Free rein: allows extensive supplementation, analogies, and associative creation.
You can set a different level of creative freedom depending on the writing topic. For example, non-fiction works should lean toward conservative, staying faithful to the source material, while literary works can give the AI more freedom to improvise.
AI model selection
When creating a writing project, you need to choose in advance the AI model that the project will use.
Different models have different characteristics, strengths, and weaknesses, and they also differ in Token consumption, writing time, and writing quality.
In general, the cheaper the model, the fewer Tokens it consumes for writing and the lower the price, but the writing quality is also relatively lower.
For example, the DeekSeek V4 model comes in a falsh version and a pro version, and the difference in their Token prices is about 3 times. The flash model writes faster and costs less, but in theory its writing quality is lower than pro (because the pro model has deep thinking capabilities).