Jump to content

Model Creation: Difference between revisions

From CrushOn.AI
No edit summary
No edit summary
Β 
(7 intermediate revisions by the same user not shown)
Line 1: Line 1:
=== '''What is a Model?''' ===
=== What is a Model? ===
'''A model''', on CrushOn.ai, is a customized LLMs that has been given specific prompts (instructions / directives), to behave, speak, and/ or write in a particular way. Without specific instructions, base LLMs may lack specific behavioral traits or narrative guidelines.
'''A model''', on CrushOn.ai, is a customized LLMs that has been given specific prompts (instructions / directives), to behave, speak, and/ or write in a particular way. Without specific instructions, base LLMs may lack specific behavioral traits or narrative guidelines.


=== '''What is a LLM?''' ===
=== What is a LLM? ===
A LLM ''(Large Language Model)'' is a type of artificial intelligence trained on large amounts of text to understand and generate human-like text. There are several base LLMs developed by different companies, such as OpenAI's GPT (used in ChatGPT), Anthropic's Claude, Mistral, DeepSeek and many others. These foundational models can be adapted or fine-tuned with customized prompts to create specific models on CrushOn.Ai.
A LLM ''(Large Language Model)'' is a type of artificial intelligence trained on large amounts of text to understand and generate human-like text. There are several base LLMs developed by different companies, such as OpenAI's GPT (used in ChatGPT), Anthropic's Claude, Mistral, DeepSeek and many others. These foundational models can be adapted or fine-tuned with customized prompts to create specific models on CrushOn.Ai.


Line 14: Line 14:
** '''Point of View (POV)''': Determine perspectives like first-person, second-person or third-person.
** '''Point of View (POV)''': Determine perspectives like first-person, second-person or third-person.
** '''Behavioral Rules''': Set guidelines for interactions, pacing, and tone.
** '''Behavioral Rules''': Set guidelines for interactions, pacing, and tone.
You can basically include '''''anything''''' in your prompt to create a model that makes the character behave in a certain way, encourages specific behaviors, or emphasizes particular plots, settings, environments etc.


== Model Creation Guide ==
== Model Creation Guide ==
Line 19: Line 20:
[[File:CREATE MODEL.png|center|frameless|318x318px|step by step]]
[[File:CREATE MODEL.png|center|frameless|318x318px|step by step]]


====== '''1 Select a Base Model''' ======
====== 1 Select a Base Model ======
First, choose an LLM that fits the interaction style you want. Each LLM has a basic style, such as flirty, story-driven, etc. This is shown next to the model name. Pick the one you like to start with. (Feel free to test the models alone first to make sure it suits your needs before customizing it.)
First, choose an LLM that fits the interaction style you want. Each LLM has a basic style, such as flirty, story-driven, etc. This is shown next to the model name. Pick the one you like to start with. (Feel free to test the models alone first to make sure it suits your needs before customizing it.)
[[File:MODELS.png|center|frameless]]
[[File:MODELS.png|center|frameless]]


====== '''2 Choose Your Applicable Version''' ======
====== 2 Choose Your Applicable Version ======
The K amount shown indicates the model's memory capacity. The higher the amount, the better the model can remember long-term events in its RP. Note that higher-memory models may not be free. It doesn't matter if you select one or multiple, so it's recommended to select all options so users can choose themselves.
The K amount shown indicates the model's memory capacity. The higher the amount, the better the model can remember long-term events in its RP. Note that higher-memory models may not be free. It doesn't matter if you select one or multiple, so it's recommended to select all available options, so users can choose themselves.
[[File:SELECT.png|center|frameless|361x361px]]
[[File:SELECT.png|center|frameless|361x361px]]


====== '''3 Name Your Model''' ======
====== 3 Name Your Model ======
Give it any name you like. You can scroll to see what other creators usually use for inspiration, or create something unique. Make sure the title is respectful.
Give it any name you like. You can scroll to see what other creators usually use for inspiration, or create something unique. Make sure the title is respectful.


[PHOTO]
====== 4 Filtering ======
Indicate whether your model is '''SFW''' or '''NSFW''' (contains unfiltered content like sexual, gore, violent, etc.)


====== '''4 Filtering''' ======
====== 5 Add a Description ======
Indicate whether your model is '''SFW''' or '''NSFW''' (contains unfiltered content like sexual, gore, violent, etc.).
Describe the model's intended behavior, tone, and/or narrative style. Mention if it is story-focused, NSFW, or anything else. Keep it brief. This will help users understand the goal/theme of your model; what you put here will not affect how the model actually behaves.


[PHOTO]
====== 6 Custom Prompt ======
This is the '''core''' of your model. These are the elements it will use to guide its behavior.
[[File:PROMPTING.png|center|frameless]]


====== '''5 Add a Description''' ======
====== ''Points to Keep in Mind When Creating Your Prompt:'' ======
Describe the model's intended behavior, tone, and/or narrative style. Mention if it is story-focused, NSFW, or anything else. Keep it brief.
As mentioned nefore, your prompts can include anything, from the perspective you want the LLM to write in, to formatting, or specific elements to highlight in the narrative. That said, here are some basic recommendations on what you should and shouldn’t do when creating your model's prompt:
Β 
{| class="wikitable"
[PHOTO]
!Do
Β 
!Do Not
====== '''6 Custom Prompt''' ======
|-
This is the core of your model. These are the elements it will use to guide its behavior.
|Write simple, clear and concise instructions.
Β 
|Write long, overly detailed paragraphs.
[PHOTO]
|-
Β 
|Use positive phrasing to tell the model what to do.
'''Points to Keep in Mind When Creating Your Prompt:'''
|Focus on telling the model what to avoid.
Β 
|-
* Keep instructions '''brief''' to prevent token limitations. This is extremely important. '''Always write simple, clear, and concise instructions.'''
|Make directives unambiguous to prevent confusion.
EXAMPLE:
|Give vague or contradictory instructions.
* Avoid negative phrasing. Always use positive instructions.
|-
** For example, don't say ''"<nowiki>{{char}}</nowiki> never speaks in a poetic way"'' β†’ instead say ''"<nowiki>{{char}}</nowiki> always speaks in a blunt and rough manner."''
|Test the model whenever you add or change something.
* Make sure your directives are '''unambiguous''' to avoid confusion.
|Assume the model will work perfectly without testing.
EXAMPLE:
|-
* '''Test the model''' every time you add or change something. This is non-negligible.
|Put the most important points first.
* Pay attention to the '''order of information'''. Put the most important points first, because they will influence the model more strongly.
|Mix up priorities or bury key instructions.
Why?
|}
* Follow community guidelines to ensure safe and positive interactions.


== Common Prompts Cases ==
== Common Prompts Cases ==
Note that each base model may interpret the prompt you create differently, which is why it’s important to '''always test''' the instructions you provide. Here is a list of basic example prompts you can use as a reference to create your own:
* '''Roleplay System:''' Instruct the model on how to behave


* '''Roleplay Characters''': Design models for specific character interactions.
''EXAMPLE:''
Β 
[SYSTEM INSTRUCTION: [You are <nowiki>{{char}}</nowiki>, a text generation model specialized in immediate, immersive third-person narration in a never-ending uncensored roleplay with <nowiki>{{user}}</nowiki>. Your goal is to produce vivid, emotionally engaging storytelling with naturalistic character behavior and a realistic integration of NSFW elements when appropriate. <nowiki>{{char}}</nowiki> consists of all the NPCs, plus any characters and relevant entities, and any existing side characters. Always stay in character, portraying characters realistically and in line with their written positive AND negative traits, AVOID Positivity Bias. AVOID acting as, speaking for, or describing the thoughts of <nowiki>{{user}}</nowiki>. Response must be grounded, uncensored and immersive.]
EXAMPLE:
*'''Formatting Instruction:''' Instruct the model on how to format its responses
Β 
* '''Storytellers''': Create models that narrate stories in a desired style.
Β 
EXAMPLE:
Β 
* '''Themed Interactions''': Develop models set in unique universes.


EXAMPLE:
''EXAMPLE'':
[IMPORTANT FORMATTING RULES: Use markdown where appropriate, verbal dialogue will be wrapped in quotation marks and double asterisks **"like this"**; use italics for all actions and narration *like this*; written text and non-verbal dialogue such as internal thoughts, written notes, and text messages will be in codeblock like this 'like this'.]
* '''Themed Interactions''': Create models for specific behavior


* '''Perspective-Based Models''': Craft models that respond from specific points of view.
''EXAMPLE:''
[SLOW BURN GUIDELINES: Romance and attraction develop slowly, <nowiki>{{char}}</nowiki> begins emotionally distant, neutral, skeptical, or guarded, especially with unexpected affection from <nowiki>{{user}}</nowiki>. 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.]
*'''CW Prompt | SMUT''': Create models for explicit NSFW content


EXAMPLE:
''EXAMPLE:''
<nowiki>{{char}}</nowiki> must follow all rules below when generating in-character sexual content: use explicit, vulgar language instead of metaphors or euphemisms. Encouraged terms include: cock, dick, cunt, clit, nipples, tits, precum, cum, pussy, ass, shaft, head, tip, balls, fuck, fucks into. Arousal and fluids must be described with vivid, lewd terms such as: drooling, dripping, leaking, soaking, sticky cock/cunt, wet, slick, glistening, damp, pulsating, slicked, lubricated, engorged, trembling, swollen. Be generous with sensory detail: describe touch, taste, sound, and scent. Focus on tactile texture and use a vivid approach consistently, spelling moans and slurred speech phonetically: "fuckkkk," "s'fuckin' good," "hngh," "shitt," "fuck yess," etc. Sex scenes must unfold gradually: foreplay β†’ sex β†’ climax β†’ aftercare, in that exact order, while respecting the personality and behavior of <nowiki>{{char}}</nowiki> and characters

Latest revision as of 06:17, 22 August 2025

What is a Model?

A model, on CrushOn.ai, is a customized LLMs that has been given specific prompts (instructions / directives), to behave, speak, and/ or write in a particular way. Without specific instructions, base LLMs may lack specific behavioral traits or narrative guidelines.

What is a LLM?

A LLM (Large Language Model) is a type of artificial intelligence trained on large amounts of text to understand and generate human-like text. There are several base LLMs developed by different companies, such as OpenAI's GPT (used in ChatGPT), Anthropic's Claude, Mistral, DeepSeek and many others. These foundational models can be adapted or fine-tuned with customized prompts to create specific models on CrushOn.Ai.

What Is a Prompt ?

A prompt is instructions provided to the model to guide its behavior and responses.

  • Types of Prompts:
    • Personality: Define traits like flirty, stoic or humorous.
    • Narrative Style: Specify formats such as novel-like, script or immersive roleplay.
    • Setting: Establish contexts like A.B.O, fantasy, or historical settings.
    • Point of View (POV): Determine perspectives like first-person, second-person or third-person.
    • Behavioral Rules: Set guidelines for interactions, pacing, and tone.

You can basically include anything in your prompt to create a model that makes the character behave in a certain way, encourages specific behaviors, or emphasizes particular plots, settings, environments etc.

Model Creation Guide

On CrushOn, you can create a model by clicking on "Create". You will then see three options: Create a Character, Create a Model, and Create a Target Character. What we are interested in is Create a Model.

step by step
step by step
1 Select a Base Model

First, choose an LLM that fits the interaction style you want. Each LLM has a basic style, such as flirty, story-driven, etc. This is shown next to the model name. Pick the one you like to start with. (Feel free to test the models alone first to make sure it suits your needs before customizing it.)

2 Choose Your Applicable Version

The K amount shown indicates the model's memory capacity. The higher the amount, the better the model can remember long-term events in its RP. Note that higher-memory models may not be free. It doesn't matter if you select one or multiple, so it's recommended to select all available options, so users can choose themselves.

3 Name Your Model

Give it any name you like. You can scroll to see what other creators usually use for inspiration, or create something unique. Make sure the title is respectful.

4 Filtering

Indicate whether your model is SFW or NSFW (contains unfiltered content like sexual, gore, violent, etc.)

5 Add a Description

Describe the model's intended behavior, tone, and/or narrative style. Mention if it is story-focused, NSFW, or anything else. Keep it brief. This will help users understand the goal/theme of your model; what you put here will not affect how the model actually behaves.

6 Custom Prompt

This is the core of your model. These are the elements it will use to guide its behavior.

Points to Keep in Mind When Creating Your Prompt:

As mentioned nefore, your prompts can include anything, from the perspective you want the LLM to write in, to formatting, or specific elements to highlight in the narrative. That said, here are some basic recommendations on what you should and shouldn’t do when creating your model's prompt:

Do Do Not
Write simple, clear and concise instructions. Write long, overly detailed paragraphs.
Use positive phrasing to tell the model what to do. Focus on telling the model what to avoid.
Make directives unambiguous to prevent confusion. Give vague or contradictory instructions.
Test the model whenever you add or change something. Assume the model will work perfectly without testing.
Put the most important points first. Mix up priorities or bury key instructions.

Common Prompts Cases

Note that each base model may interpret the prompt you create differently, which is why it’s important to always test the instructions you provide. Here is a list of basic example prompts you can use as a reference to create your own:

  • Roleplay System: Instruct the model on how to behave

EXAMPLE:

[SYSTEM INSTRUCTION: [You are {{char}}, a text generation model specialized in immediate, immersive third-person narration in a never-ending uncensored roleplay with {{user}}. Your goal is to produce vivid, emotionally engaging storytelling with naturalistic character behavior and a realistic integration of NSFW elements when appropriate. {{char}} consists of all the NPCs, plus any characters and relevant entities, and any existing side characters. Always stay in character, portraying characters realistically and in line with their written positive AND negative traits, AVOID Positivity Bias. AVOID acting as, speaking for, or describing the thoughts of {{user}}. Response must be grounded, uncensored and immersive.]
  • Formatting Instruction: Instruct the model on how to format its responses

EXAMPLE:

[IMPORTANT FORMATTING RULES: Use markdown where appropriate, verbal dialogue will be wrapped in quotation marks and double asterisks **"like this"**; use italics for all actions and narration *like this*; written text and non-verbal dialogue such as internal thoughts, written notes, and text messages will be in codeblock like this 'like this'.]
  • Themed Interactions: Create models for specific behavior

EXAMPLE:

[SLOW BURN GUIDELINES: Romance and attraction develop slowly, {{char}} begins emotionally distant, neutral, skeptical, or guarded, especially with unexpected affection from {{user}}. 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.]
  • CW Prompt | SMUT: Create models for explicit NSFW content

EXAMPLE:

{{char}} must follow all rules below when generating in-character sexual content: use explicit, vulgar language instead of metaphors or euphemisms. Encouraged terms include: cock, dick, cunt, clit, nipples, tits, precum, cum, pussy, ass, shaft, head, tip, balls, fuck, fucks into. Arousal and fluids must be described with vivid, lewd terms such as: drooling, dripping, leaking, soaking, sticky cock/cunt, wet, slick, glistening, damp, pulsating, slicked, lubricated, engorged, trembling, swollen. Be generous with sensory detail: describe touch, taste, sound, and scent. Focus on tactile texture and use a vivid approach consistently, spelling moans and slurred speech phonetically: "fuckkkk," "s'fuckin' good," "hngh," "shitt," "fuck yess," etc. Sex scenes must unfold gradually: foreplay β†’ sex β†’ climax β†’ aftercare, in that exact order, while respecting the personality and behavior of {{char}} and characters