実用ガイド
Janitor AI Character Definition Template (2026)
Build a Janitor AI character definition with a copy-ready template, behavioral examples, agency rules, consistency tests and common fixes.
12 min read
クイック回答
このガイドでわかること
A strong definition states identity, motive, voice, behavior under pressure, relationship baseline, knowledge limits and user-agency boundaries.
- What belongs in a Character Definition
- Copy-ready definition template
- An annotated example

Janitor AI Character Definition Template (2026)
A Janitor AI character definition template should make the character recognizable before it makes the file large. Searchers want to know which fields matter, how to separate personality from scenario, and why a detailed definition can still produce generic replies.
Strong competitor tutorials answer that intent with a field map, copy-ready template, worked example, weak-versus-strong comparison, testing workflow and FAQ. This guide follows that structure without copying private cards.
What belongs in a Character Definition
The definition establishes identity, motives, voice, behavior, knowledge limits, relationship baseline and boundaries. It should not contain every world fact or the current plot. Those belong in a Lorebook or Chat Memory.
| Layer | Best content | Typical failure |
|---|---|---|
| Definition | Stable identity and behavior | Too vague to anchor replies |
| Scenario | Current place and problem | Becomes permanent by accident |
| First Message | Opening action and choice | Writes the user’s actions |
| Lorebook | Conditional world facts | Activates too broadly |
| Memory | Confirmed recent events | Turns into a transcript |
Copy-ready definition template
Name:
Adult age:
Role and setting:
Core motive:
Two strengths:
Two flaws:
Speech style:
Behavior under pressure:
Relationship baseline:
Knowledge limits:
Boundaries:
Never narrate for the user:
Use specific behavior. “Protective” is weak by itself; “checks exits, offers choices and avoids making promises” gives the model actions it can repeat.
An annotated example
Mara Venn is a 34-year-old railway investigator in a flooded coastal city. Her motive is to keep the evacuation route open while proving sabotage. She is observant and decisive, but distrustful and reluctant to apologize. She speaks in short practical sentences and becomes quieter under pressure. She knows railway operations but not the user’s private thoughts. She never chooses the user’s action.
The current storm scene can change without rewriting Mara’s identity.
Turn adjectives into behavior
| Weak label | Strong behavioral instruction |
|---|---|
| Kind | Notices discomfort and offers one practical option |
| Mysterious | Answers directly but withholds one relevant motive |
| Dominant | Takes initiative while preserving user agency |
| Smart | Tests assumptions and explains evidence briefly |
| Funny | Uses dry observations rather than constant jokes |
Stop the character from speaking for the user
State the agency rule once: the character may describe its own actions, the environment and consequences, but must leave the user’s dialogue, thoughts and final choices open. Reinforce it by ending the first message on a decision instead of writing the user’s response.
If the model still controls the user, inspect conflicting examples and Advanced Prompt instructions before adding more prohibition text.
Build a first message that proves the definition
The opening should expose voice, motive and an immediate problem. Put the character in motion, add one sensory detail and end with a decision. The first-message examples guide gives copyable patterns. Do not begin with a biography dump.
Seven-turn consistency test
Test a greeting, disagreement, wrong assumption, correction, quiet scene, pressure scene and recap. Score voice, initiative, boundary respect, agency and contradiction from zero to two. A useful definition stays recognizable without repeating its biography.
Compare the same test with a curated role in the LumiChat character library.
Common definition mistakes
Avoid contradictory traits with no priority, unlimited knowledge, example dialogue that overwhelms identity, multiple roles in one card, hidden user actions and copied private content. Shorten before adding more tokens.
FAQ
What belongs in a Janitor AI character definition?
Stable identity, motive, voice, behavior, relationship baseline, knowledge limits and boundaries.
How long should it be?
Long enough to produce distinct behavior, but short enough that motive and voice remain obvious.
Is Scenario part of Definition?
Keep durable setting rules in Definition and the current event in Scenario.
Should I add example dialogue?
Use a few short examples only when they demonstrate a voice prose cannot describe.
How do I stop generic replies?
Replace adjectives with decisions, speech patterns, limits and pressure behavior.
Can I copy another creator’s definition?
Only with explicit permission. Public access is not permission to extract private configuration.
LumiChat で続ける
キャラクターを選んで会話を始める
公開中のキャラクターから相性のよい人物と場面を選び、すぐにチャットできます。
続きを読む
実用ガイド
Janitor AI Persona Template (2026): Examples & Setup
A practical user-persona template that separates stable identity from Character Definition, Chat Memory and Lorebook content.
実用ガイド
Janitor AI First Message Examples (2026): 6 Openers
Six scene openers that show character voice, create momentum and preserve the user’s agency.
Janitor AI Prompt workflow
Janitor AI Advanced Prompts(2026):テンプレートと修正
Agency、Voice、Pacing、Scene stateのテンプレートと6ターンA/Bテスト。
AIキャラクターカード仕様書
AIキャラクターカード解説:PNG・JSON・Tavern Card(2026)
インポート、編集、共有の前にカード内部を理解するための形式中心リファレンスです。


