Tokens and context length

Character text, system prompts, lorebook entries, user persona, chat history, and the current message all consume context. Output also needs reserved space.

Definition tokens are only the fixed cost; conversation history grows over time.

A concise, well-structured definition can outperform repetitive long text. Large cards support complex worlds but require more context and inference cost.

Choose based on model context, expected chat length, and lorebook behavior.

Trigger only necessary lore, summarize long chats, and remove repeated or conflicting instructions. Avoid copying the same background into multiple fields.

When early details are forgotten, inspect truncation and prompt priority before adding more permanent text.