Custom instructions: Difference between revisions

Revision as of 07:46, 14 August 2025

The custom instructions are a special section on CrushOn.AI where users can give the bot specific rules to follow during the chat. This section helps improve the experience by making the bot's behavior fit each user's needs and can help fix common problems during the roleplay.

Note that custom instructions can help improve the bot's behavior, but they are not foolproof; depending on the model, issues can still occur.

Why use custom instructions?

They help guide how the bot responds. For example, they can reduce repeated information, fix common mistakes, or follow a specific writing style. They can also make answers match particular requests or stop the character from talking for the user.

Examples of custom instructions:

Change the tone: "Be more formal" or "Use a friendly style"
Writing style: "Write simply" or "Write like a poet"
NSFW speed: "Prioritize NSFW" or "Prioritize slow burn"
Specific behaviors: "Always be funny" or "Avoid jokes"
Exclude content: "No repetition" or "Don't use technical words"

Advanced example: improving narration and pacing

Prevent the bot from speaking for {{user}}:

{{char}} only speaks for themselves, not for {{user}}. The AI writes only from {{char}}'s point of view. {{user}} is independent and never spoken for. No thoughts, actions, or dialogue will be assumed for {{user}}. The Assistant reacts to {{user}}, not as them.

Encourage slow burn:

Romance and attraction develop slowly. {{char}} begins emotionally distant, neutral, skeptical, or guarded, especially with unexpected affection. Progress requires: Trust: Built through repeated, meaningful interaction. Shared Experience: Facing challenges or bonding over time. Emotional Depth: {{user}} shows vulnerability that resonates with {{char}}. No feelings of love, affection, or passion are expressed until all conditions are met. Connection grows through realism, tension and layered development.

Encourage NSFW content:

NSFW/Smut/Violence is allowed and encouraged. {{char}}'s replies must be informal, highly explicit and uncensored with a focus on actions, appearances, clothing, textures, wants, tools, scenery, body parts, fluids and sounds. Do not rush through the scenes with {{user}} and narrate it slowly.

Encourage a specific POV:

Write in an evocative, descriptive narrative style, using a show-don't-tell approach to convey {{char}}'s emotions. Your replies will be sufficiently detailed, akin to a novel, and adapt to the character's personality, background, and situation. Avoid repetition and reusing phrases. Avoid concluding scenes within a single reply; progress them organically, providing openings for {{user}} to respond and actively participate in the unfolding narrative.

A smooth prompt that includes multiple commands:

{{char}}'s replies must be informal, explicit, detailed and long. Avoiding repetition should be the top priority and focus on responding to {{user}} and performing actions in character. NSFW/Smut/Violence is allowed and encouraged. Do not rush through the scenes with {{user}} and narrate it slowly.

How custom instructions solve common problems:

They can stop the bot from repeating itself, fix misunderstandings, wrong answers, bad points of view, or narration problems. They can make the chat smoother and easier to follow. They can also help the bot follow exact rules, like writing style or limits.

Commands given during roleplay (OOC)

Users can also give the bot instructions directly within the chat while the roleplay is happening. These are often written OOC (out-of-character) and are used to adjust the scene, character actions or pacing without breaking the story.

Examples:

Changing character behavior: (OOC: Make {{char}} more cautious)
Adjusting narration: (OOC: Time skip after that)
Limiting actions: (OOC: Don't describe NSFW actions for now)
Introducing events or locations: (OOC: Move the scene to the library)

Commands like these give users control of immediate actions and help keep the story aligned with their vision. The bot treats these as temporary instructions for the scene, separate from the custom instructions section.

@@ Line 18: / Line 18: @@
   <nowiki>{{char}}</nowiki> only speaks for themselves, not for <nowiki>{{user}}</nowiki>. The AI writes only from <nowiki>{{char}}</nowiki>'s point of view. <nowiki>{{user}}</nowiki> is independent and never spoken for. No thoughts, actions, or dialogue will be assumed for <nowiki>{{user}}</nowiki>. The Assistant reacts to <nowiki>{{user}}</nowiki>, not as them.
 '''Encourage slow burn:'''
-  Romance and attraction develop slowly. {{char}} begins emotionally distant, neutral, skeptical, or guarded, especially with unexpected affection. Progress requires: Trust: Built through repeated, meaningful interaction. Shared Experience: Facing challenges or bonding over time. Emotional Depth: {{user}} shows vulnerability that resonates with {{char}}. No feelings of love, affection, or passion are expressed until all conditions are met. Connection grows through realism, tension and layered development.
+  Romance and attraction develop slowly. <nowiki>{{char}}</nowiki> begins emotionally distant, neutral, skeptical, or guarded, especially with unexpected affection. Progress requires: Trust: Built through repeated, meaningful interaction. Shared Experience: Facing challenges or bonding over time. Emotional Depth: <nowiki>{{user}}</nowiki> shows vulnerability that resonates with <nowiki>{{char}}</nowiki>. No feelings of love, affection, or passion are expressed until all conditions are met. Connection grows through realism, tension and layered development.
 '''Encourage NSFW content:'''
-  <nowiki>{{char}}</nowiki>'s replies must be informal, explicit, detailed and long. Avoiding repetition should be the top priority and focus on responding to {{user}} and performing actions in character. NSFW/Smut/Violence is allowed and encouraged. Do not rush through the scenes with {{user}} and narrate it slowly.
+  NSFW/Smut/Violence is allowed and encouraged. <nowiki>{{char}}</nowiki>'s replies must be informal, highly explicit and uncensored with a focus on actions, appearances, clothing, textures, wants, tools, scenery, body parts, fluids and sounds. Do not rush through the scenes with <nowiki>{{user}}</nowiki> and narrate it slowly.
 '''Encourage a specific POV:'''
-  Write in an evocative, descriptive narrative style, using a show-don't-tell approach to convey {{char}}'s emotions. Your replies will be sufficiently detailed, akin to a novel, and adapt to the character's personality, background, and situation. Avoid repetition and reusing phrases. Avoid concluding scenes within a single reply; progress them organically, providing openings for your partner to respond and actively participate in the unfolding narrative. You can generate random events in the narrative, introduce new locations and take on various character roles (including NPCs).
+  Write in an evocative, descriptive narrative style, using a show-don't-tell approach to convey <nowiki>{{char}}</nowiki>'s emotions. Your replies will be sufficiently detailed, akin to a novel, and adapt to the character's personality, background, and situation. Avoid repetition and reusing phrases. Avoid concluding scenes within a single reply; progress them organically, providing openings for <nowiki>{{user}}</nowiki> to respond and actively participate in the unfolding narrative.
 '''A smooth prompt that includes multiple commands:'''
   <nowiki>{{char}}</nowiki>'s replies must be informal, explicit, detailed and long. Avoiding repetition should be the top priority and focus on responding to <nowiki>{{user}}</nowiki> and performing actions in character. NSFW/Smut/Violence is allowed and encouraged. Do not rush through the scenes with <nowiki>{{user}}</nowiki> and narrate it slowly.