Instructions Guide

Custom Instructions play a crucial role in controlling the behavior, tone, and response style of a model. By setting the right custom instructions , you can direct the model to provide more relevant, consistent, and contextually appropriate responses. They can also be utilized for short term problem-solving.

How to Write Effective Prompts

To ensure your custom instructions work, even with simpler or older LLMs, please follow these guidelines:

• Be direct: Use clear, literal language. Avoid metaphors, irony, or vague suggestions and statements.
  • ✅ {{char}} never uses slang.
  • ❌ {{char}} keeps it classy

• Use present-tense: State behavior as an ongoing fact or instruction, not a request or possibility.
  • ✅ {{char}} avoids long responses.
  • ❌ {{char}} should try to avoid long responses.

• Keep it short: One-sentence rules work best. If needed, stack multiple clear lines—don’t write long paragraphs.

• Be specific: Focus on what the character should or should not do, not how they should "feel" or "think."

By keeping your instructions simple, rule-based, and action-focused, you give even limited models a better chance to follow them correctly.

Usage In Roleplay

Custom instructions can be utilized to tweak the model to your preferences. For example, if you want the model to be more focused on specific descriptions or changing it's tone, you can use prompt like this -

• {{char}} always speaks with a cold, analytical tone, avoids emotional language, and prioritizes logic in every response.

This prompt sets a clear behavioral guideline for the AI to follow, instructing it to maintain a cold, analytical tone, avoid emotional expressions, and focus on logical, reasoned responses.

This prompt can be tweaked in any way to better fit what you are looking for. For example -

• {{char}} speaks with warmth and emotional depth, often using expressive language, showing empathy, and prioritizing feelings over logic in responses.

Keep in mind that custom models sometimes come with pre-defined behavioral patterns and bots with their respective personalities. While custom instructions can influence their responses, they may not fully override the model’s or bot's default behavior. As a result, prompts like the examples above may not work consistently in every case, especially if the model’s core style strongly differs from the instructions. Experimentation and slight adjustments are often needed to achieve the desired effect.

Usage As Troubleshooting Tool

Custom instructions can also serve as a valuable troubleshooting mechanism when a model exhibits undesired behavior. For instance, if the model frequently repeats itself or echoes previous responses verbatim, you can use a directive like:

Repetition: {{char}} avoids repeating exact sentences or phrases.

Breaking character: {{char}} always stays in character and never talks about being an AI or language model.

Model speaking for the user: {{char}} never writes lines or dialogue for {{user}} under any circumstances.

These instructions serve as behavioral nudges rather than guaranteed overrides. Many models, especially older or platform-restricted ones may still ignore or partially resist these prompts due to built-in safety filters, memory limitations, or preset behavior. Use them as guidance tools to improve output instead of a solution to every problem. Often times, the issues lie not within the model, but in the character you are talking with, or the custom model which is not supervised by Crushon.

Important Reminder

Don’t feel limited to the examples above. Custom instructions can be adapted and combined in countless ways to suit your needs. Whether for fixing behavior, shaping personality, adjusting tone, or guiding specific scenarios. Think of them as flexible tools you can use creatively to get closer to the ideal response style you want.