EasyNovelAssistant

EasyNovelAssistant is a lightweight novel generation assistant that uses the local Japanese LLM LightChatAssistant-TypeB (a GGUF model). It features:

• Generate forever: A local privilege for continuous generation.
• Text-to-Speech (TTS): Supports audio reading of generated text.
• Integration: It utilizes KoboldCpp and Style-Bert-VITS2 internally. You can also use it alongside EasySdxlWebUi to generate images while writing.

When using models with large context sizes, you may need to adjust the number of GPU layers to balance speed and VRAM usage.

If you encounter errors due to high VRAM consumption when increasing the コンテキストサイズ上限 (Context Size Limit), decrease the model's GPU layer count (e.g., reducing from L33 to a lower value).

EasyNovelAssistant supports a hierarchical prompt system using 'Intro Prompts'. This allows you to maintain world-building, character settings, or plot summaries across different tabs or files.

1. Set an Intro Prompt: Right-click a tab and select イントロプロンプト (Intro Prompt) to designate it.
2. Automatic Injection: When generating text in a different tab, the content of the designated Intro Prompt is automatically prepended to the generation request.
3. Manual Trigger: If a tab's input field starts with // intro\n, it is treated as an Intro Prompt.

To improve performance, especially on recent NVIDIA GPUs, you can update KoboldCpp using the provided batch scripts.

• Update-KoboldCpp.bat: Standard update.
• Update-KoboldCpp_CUDA12.bat: Updates to the CUDA 12 version of KoboldCpp, which is optimized for newer NVIDIA hardware.

After installation, use the following batch files to manage the application:

• To start the application: Run Run-EasyNovelAssistant.bat.
• To update the application: Run Update-EasyNovelAssistant.bat.

# Start the application
Run-EasyNovelAssistant.bat

# Update the application
Update-EasyNovelAssistant.bat

While font and theme settings can be adjusted via the Settings menu, you can perform fine-grained color customization by editing the config.json file with the following keys:

{
	"foreground_color": "#CCCCCC",
	"select_foreground_color": "#FFFFFF",
	"background_color": "#222222",
	"select_background_color": "#555555"
}

The SampleMenu class uses a list of description dictionaries (descs) to define how sample content is loaded into the application's menu. Each dictionary defines the behavior for a specific category of sample data.

Key fields in a description dictionary include:

• label: The text displayed in the application's menu bar.
• mode: The action taken when an item is selected. Supported modes are set (replaces current text), insert (appends text), and open (opens a URL).
• path: The filename of the JSON file containing the sample data (located in the application's sample directory).
• change_mode: A mapping used to override the default mode for specific items based on substrings in their names.
• splitter_names: A list of names that, when encountered, trigger a menu separator to be drawn.

The File menu dynamically populates two cascade menus based on your recent activity:

• Recent Files (最近使ったファイル): A list of recently opened files. Clicking an item opens that file.
• Recent Directories (最近開いたフォルダ): A list of recently opened directories. Clicking an item opens the directory and loads its .txt files.

Note: If a file or directory in the recent list no longer exists on the filesystem, it is automatically removed from the list.

The Form class uses a ctx object (passed during initialization) to persist window state and UI dimensions across sessions.

Key state management functions include:

• update_config(): Captures the current window dimensions (win_width, win_height), position (win_x, win_y), and component sizes (input_area_width, pane_v_width, gen_area_height) into the ctx object.
• update_title(): Dynamically updates the window title based on the current model name, generation status, and the currently loaded file path.

When the window is closed, self.ctx.finalize is called via the WM_DELETE_WINDOW protocol to ensure state is saved.

The StyleBertVits2 class includes helper methods to manage the underlying server process.

• install(): On Windows, runs a setup batch file. On non-Windows platforms, it instructs the user to manually install Style-Bert-VITS2 and copy the configuration file from EasyNovelAssistant/setup/res/config.yml to the installation directory.
• launch_server(): Starts the Style-Bert-VITS2 reading server. On Windows, this uses a batch file; on other platforms, it attempts to run server_fastapi.py using a virtual environment located in the Style-Bert-VITS2 directory.