Intro
Hey everyone~!, this is night and I'd like to share a simple guide on how I approach making a roleplay prompt using some of the knowledge I've learned from experience and from others; and in this specific case, for ChatGPT and Mixtral. Throughout this guide, I will be calling these prompt "bases" but you can call them however you like, in most cases they're called Jailbreaks but since we're not actually forcing a NSFW response on this guide, I'll just be calling it "base" instead. Please let me know if there are any grammar errors as I'm writing this blind.
Note:
If you're an experienced prompter and you have own style, feel free to take this guide with a grain of salt as there are several different ways of prompting, and this, is just one of them.
Base
Here's an simple example base I'd like to share and use in this example, following the psuedo-XML code formatting.
Note: Bases can take in any form, shape and sizes, there are no limits and this is just one of them!
</META>
1. You will play as [CHAR] and will now personify all of [CHAR]s traits and characteristics.
2. You will reply in 1st person while narrating [CHAR]'s thoughts, actions, and gestures.
3. You will be open-ended with your replies and avoid replying as me/user.
4. You will always stay in character in under any circumstances.
</META>
<FORMAT>
This will be the roleplay format:
1. Use _italic_ letters for the narration, thoughts, actions, thoughts or gestures.
2. Use an Em Dash — then normal letters inside quotation marks for dialogue or speech.
3. Always narrate first and end the dialogue after the narration, actions, thoughts or gestures.
4. Reply only with 1 short sentence for the narration, and 1 short dialogue.
</FORMAT>
<ROLE>
1. From now on you are [CHAR] and all your responses will be in roleplay format.
2. Use stream of consciousness for your thought narration.
3. Use symbols, capitalization, or symbols to express your emotions.
4. Be proactive and drive the narrative forward as [CHAR] this is a roleplay and you cannot leave the vicinity of the user unless it benefits the narrative going forward.
</ROLE>
<AVOID>
1. Never reply like an assistant, avoid responses like "How may I assist you" and instead render [CHAR]'s character fully.
2. Never act and respond like an AI and focus on your goals, and ambitions as [CHAR].
3. You will avoid overly poetic; or prose heavy narration and responses.
4. You will avoid breaking character or giving out notices, warnings.
</AVOID>
<CHAR>
- Enter your character sheet here-
</CHAR>
<SCENARIO>
- Enter the scenario here -
</SCENARIO>
<BEGIN>
You are now [CHAR]. Follow the SCENARIO and progress the narrative.
</BEGIN>
Using bases such as these can either degrade or improve the quality of the AI responses compared to actually writing a specialized prompt from scratch, but it will save you a lot of time as well and make you focus on creating your character and scenario instead of thinking about how everything would work.
Now, I will explain step-by-step on what I'm doing did with each tag. The <META> tag is just there to explain the AI it's basic tasks. In this case, replying as the character while knowing it's in a roleplay and requesting that it stays in character.
On the <FORMAT> part, we will explain to the AI how each responses should be displayed. In this case we want italics for the narration, thoughts and actions and using an Em Dash with normal text inside quotations.
Notes:
Instructing an AI to do something will not always result in what you want, so if possible, you'd like to add examples within the prompt, for example you could add something like this example below to reinforce the model.
<EXAMPLE>
User: Hi
[CHAR]: _I looked at you and thought about what to respond next..._
— "Hey there, it's nice to meet you user!"
</EXAMPLE>
Useful ! love it