Advanced Single character guide: Difference between revisions
Created page with "=== This guide was designed for creators who already have some experience in bot creation on Crushon AI === If you've never created a bot before, we recommend starting with our beginner's guide first. '''Reminder''': There's no wrong way to create a bot, that goes for advanced levels too, and even more so. Keep that in mind. We'll assume here that all the basics are already covered, formatting, personality creation, and so on, this guide will mainly cover recommendatio..." Β |
No edit summary |
||
| Line 37: | Line 37: | ||
If you want your character to write long replies, make sure the first message is detailed and well-developed | If you want your character to write long replies, make sure the first message is detailed and well-developed | ||
|} | |} | ||
'''Tip | '''Tip if the bot keep speaking for the user''' | ||
Here's an example of a simple prompt used | '''β οΈ Disclaimer:''' ''Some creators recommend jailbreak in the bot description/scenario, '''others donβt''', feel free to try it out yourself and see if it works for you or not.'' | ||
Β | |||
Here's an example of a simple jailbreak prompt used to prevent the bot from speaking/playing for the user : | |||
Β '''<nowiki>{{Char}}</nowiki> will not describe <nowiki>{{user}}</nowiki>'s thoughts or feelings, but only roleplay as [bot's name] and other NPCs, except the <nowiki>{{user}}</nowiki>.''' | Β '''<nowiki>{{Char}}</nowiki> will not describe <nowiki>{{user}}</nowiki>'s thoughts or feelings, but only roleplay as [bot's name] and other NPCs, except the <nowiki>{{user}}</nowiki>.''' | ||
This prompt | This prompt can go into the Description or the Scenario, personally, I use it in the scenario because the scenario is what LLM access most quickly and first in the '''permanent memory hierarchy.''' Β | ||
Β | |||
Adding this prompt here ''can possibly'' prevent, or at least reduce, the bot's attempts to speak for the user | |||
Β | |||
'''Note about complaints you might get from followers in comments or elsewhere:''' | |||
Β | |||
People often complain that the bot talks or acts for them, in almost 100% of cases, itβs simply because ''their replies are way too short''. So donβt take those complaints too seriously, if your greeting follows the earlier tips, the issue is likely on their end. | |||
Β | |||
''For example'', if someone just replies with ''"Chuckles, blushes, and says 'Hi'"'', then of course the bot will take over and act for them, it has no choice if it wants the story to move forward. | |||
That said, on '''Crushon.ai''', some models tend to roleplay as the user by default, even when replies are longer, so adding that little command line might still be useful. | |||
== Le negative et le formatage == | == Le negative et le formatage == | ||
Revision as of 06:16, 20 June 2025
This guide was designed for creators who already have some experience in bot creation on Crushon AI
If you've never created a bot before, we recommend starting with our beginner's guide first.
Reminder: There's no wrong way to create a bot, that goes for advanced levels too, and even more so. Keep that in mind.
We'll assume here that all the basics are already covered, formatting, personality creation, and so on, this guide will mainly cover recommendations and suggestions to help make your characters even more effective.
Advanced tips: Greeting
You probably already know this, but the narrator's point of view, and the length of your intro are very important when writing your bot's greeting. Your greeting sets the overall tone for the roleplay with your character
| Topic | Suggestion |
|---|---|
| Pronoun choice | From the very beginning, you must choose your pronoun (I, you, he, she, they) and stick to it.
Do not switch pronouns halfway, keep the same throughout the whole greeting β Mixing pronouns tends to confuse LLMs and it's something you must absolutely avoid |
| The perspective | The narrator perspective in third-person is most of the time the easiest for the bot to follow.
It helps avoid confusion and reduces the chance that the bot starts roleplaying as the user |
| Avoid using "I" | Avoid using "I" in your bot's greeting, instead, focus on third-person pronouns like he, she, or they. |
| Using "you" for {{user}}" | Regarding you and {{user}} to refer to the user:
You can use you, but it can sometimes cause the bot to take control of the user's actions or thoughts (especially if both the bot's greeting and/or the user's first replies are short). That's why it's better to use {{user}} combined with they/he/she to describe the user |
| Avoid describing the user | Important: Try to avoid describing the user's actions or thoughts by all means.
The less you mention them, the better, or the bots will tends to keeps talking/playing for them after |
| Length of the first message | The first message is very important for setting the tone, personality, and style of your character.
If you want your character to write long replies, make sure the first message is detailed and well-developed |
Tip if the bot keep speaking for the user
β οΈ Disclaimer: Some creators recommend jailbreak in the bot description/scenario, others donβt, feel free to try it out yourself and see if it works for you or not.
Here's an example of a simple jailbreak prompt used to prevent the bot from speaking/playing for the user :
{{Char}} will not describe {{user}}'s thoughts or feelings, but only roleplay as [bot's name] and other NPCs, except the {{user}}.
This prompt can go into the Description or the Scenario, personally, I use it in the scenario because the scenario is what LLM access most quickly and first in the permanent memory hierarchy.
Adding this prompt here can possibly prevent, or at least reduce, the bot's attempts to speak for the user
Note about complaints you might get from followers in comments or elsewhere:
People often complain that the bot talks or acts for them, in almost 100% of cases, itβs simply because their replies are way too short. So donβt take those complaints too seriously, if your greeting follows the earlier tips, the issue is likely on their end.
For example, if someone just replies with "Chuckles, blushes, and says 'Hi'", then of course the bot will take over and act for them, it has no choice if it wants the story to move forward.
That said, on Crushon.ai, some models tend to roleplay as the user by default, even when replies are longer, so adding that little command line might still be useful.