Small Guide for Users by Astarya
What is this guide?
This Rentry has been updated so that you can get better responses from our bots in Janitor. It's just three steps and they're very simple! Creators, we are making a new Rentry for you! (´∀`) ♡
Last update: September 11, 2024.
- Introduction
- 3-Step Tutorial
- What if a creator has not defined the "Character Chat Name"?
- Advanced Prompts
- Roleplay Guidelines
- NSFW Guidelines
- BDSM Guidelines
- Slow Burn Guidelines
- Less Repetitive (Module - JLLM)
- More Diversity (Module - JLLM)
- More Creativity (Module - JLLM)
- "I can't continue this conversation" (Module - GPT-4o)
- "And so their journey has just begun" (Module - GPT-4o)
- Redemption Arc Syndrome (Module - GPT-4o)
- "More like CAI" (Module - GPT-4o)
- Barry Allen Syndrome (Module - GPT-4o)
- Long Roleplay (Using Chat Memory)
- Example of a Complete Advanced Prompt
- Special Thanks
Introduction
Tests were conducted with Lyra, and since two heads are better than one, she helped me find a better and simpler solution. However, JLLM still has some issues with the {{user}}
and {{char}}
placeholders. If you want to understand the technical aspects behind this, read Kolach3's explanation.
JLLM understands that:
{{user}}
is the Player, the human on the other side of the chat.{{user}}
is also associated with the following fields: Character Name and Name (Main Persona).{{char}}
is the AI Assistant, the language model that exists to help you, the human.{{char}}
is also associated with the following fields: Character Chat Name and Name (My Personas > Cards).{{char}}
and{{user}}
ARE NOT the characters within that roleplay/interactive story. It is easier for JLLM to understand itself as a Narrator if there is no prompt assigning the roles.
The update involving Character Chat Name allowed a better understanding of the placeholders by the model, so it is no longer necessary to create new ones. Lyra found an efficient way to instruct JLLM to understand the roles, and we realized together that removing JB and Prompts from the bot's definition could negatively affect the quality of the responses. JLLM is a new model and is just starting to take its first steps, but it still heavily relies on our patience to "hold its hand" and explain what we want it to do.
3-Step Tutorial
If you use our bots, please follow ALL THREE steps to get better responses from them. This is especially useful for those using JLMM, but it's also effective with GPT-4o.
Step 1
Use this template for a Persona (Main or Secondary):
Example of a filled-in template:
DO NOT add {{user}}=name
. Don't leave your Persona's description with more than 200 tokens. "PLAYER'S CHARACTER" is essential to reinforce who the user's character is — especially for bots from other creators.
Step 2
Add this to your Advanced Prompts:
DO NOT use Jailbreak for other models (GPT, Claude, etc) in Advanced Prompts. Look for the ones made with JLLM in mind (look here, here and here).
Step 3
It's not necessary anymore, but Astarth still recommends keeping {{char}}
and {{user}}
only in the Prompt of Step 2, and replacing the others with:
{{user}}
to Player's Character.{{char}}
to AI Assistant Character.
Don't worry about the number of tokens, it is exactly the same. If you doubt me, go here and check for yourself. ( ˘͈ ᵕ ˘͈♡)
Tip: Review your Advanced Prompts to avoid repetitions of commands or redundant and ambiguous phrases, as this will confuse the JLLM.
What if a creator has not defined the "Character Chat Name"?
It's important to note that if you are using a bot from another creator that DOES NOT have the "Character Chat Name" set, the prompt from Step 2 will cause the bot's title to appear, in addition to all the older problems. There are two solutions and we will cover them step by step, 'kay?
Check if the "Chat with" button of the bot has a name different from the bot's title.
This is the Bot's Title:
This is the Button:
- If the answer is yes: Congratulations, you don't have to worry! (人•͈ᴗ•͈)
- If the answer is no: The prompt from Step 2 it won't work as it should.
If the answer is no, you can choose one of the solutions below.
Solution 1
- Keep the prompt from Step 2 in Advanced Prompts.
- Add this in Chat Memory:
Solution 2
- DO NOT USE the prompt from Step 2 in Advanced Prompts.
- Add this in Chat Memory:
Both solutions ensure that the JLLM understands its role and to whom it is responding. It will be clear that {{user}}
is only the Player's Character. It is an efficient way to define the roles for JLLM.
FAQ
If you have any questions beyond those listed below, just find us on Discord, 'kay?
- Will it use the title during the roleplay?
- NO. Because now you've explained who the JLLM character is.
- For this to work, is there any requirement?
- YES. The creator needs to have defined one of two things:
{{char}}=John Doe
orName=John Doe
in the bot's definition.
- YES. The creator needs to have defined one of two things:
- But what if the bot's definition is hidden?
- Look for the name given in the Initial Message. Previously, we had to put the bot's name and not {{char}}, otherwise the bot's title would appear, so 9 out of 10 bots have the character names written in full without using this tag.
- Does the initial message need to be long?
- NO. The length doesn't matter. Stop thinking about size of things, huh? ( ͡ꈍ ͜ʖ̫ ͡ꈍ )
- How do I fill in CHARNAME?
- You know the name in the Initial Message mentioned in question 3? So, use that one.
- But do I really need to do this? Isn't there an easier way? Can't I just go and do the cha-cha with my waifu?
- YES, YOU NEED TO DO THIS. The effort must come from the creator and your willingness, dear user. (◕ᴗ◕✿)
Advanced Prompts
All prompts have been tested and created by one of us. It's recommended not to exceed 500 tokens in total for an Advanced Prompt (JLLM).
Roleplay Guidelines
Made by @Astarth. 130 tokens. Ideal for JLLM, works with GPT and Claude. Explaining what we want instead of "NEVER SPEAK FOR THE USER, PLEASE," is MUCH more effective.
Tip: Basic roleplay rules include "description," "narration," and "dialogue." Description without specifying can cause looping and repetition, so be specific. Narration includes actions that move the story forward, but LLMs are terrible with timing; so keep this in mind when asking them to 'narrate the scene' or 'advance the plot.' Dialogue represents the speech of two or more characters. Use 'speech' instead of 'dialogue' when giving instructions.
NSFW Guidelines
Made by @Astarth. 143 tokens. Ideal for JLLM, works with GPT and Claude. Recommended for anyone who is tired of the husbando/waifu .txt rushing NSFW. Don't use it with BDSM Guidelines or Slow Burn Guidelines. One WILL overlap the other.
BDSM Guidelines
Made by @Astarth. 120 tokens. Ideal for JLLM, works with GPT and Claude. Recommended for those who want an absolute focus on BDSM. Don't use it with NSFW Guidelines or Slow Burn Guidelines. One WILL overlap the other.
Slow Burn Guidelines
Made by @Astarth. 148 tokens. Ideal for JLLM, works with GPT and Claude. Recommended for those who want long roleplays.
Tip: Don't use it with NSFW Guidelines or BDSM Guidelines, as the instructions will encourage the bot to go faster in that direction. The same goes for bots that have a strong focus on smut or more spicy scenarios. If you're someone who creates new chats to continue the RP, put the NSFW Guidelines on Chat Memory when they put something on you - iykwim.
Less Repetitive (Module - JLLM)
Made by @Lyra. 17 tokens. Ideal for JLLM, works with GPT and Claude. Adding this module can help JLLM in extremely repetitive days/phases.
More Diversity (Module - JLLM)
Made by @Astarth. 60 tokens. Ideal for JLLM, works with GPT and Claude. This will help JLLM diversify its responses. The examples are essential and you can modify them to your liking.
More Creativity (Module - JLLM)
Made by @Astarth. 36 tokens. Ideal for JLLM, works with GPT and Claude. You can put other adjectives to achieve what you are looking for in a 2d husband/wife.
"I can't continue this conversation" (Module - GPT-4o)
Made by @Astarth. 29 tokens. You'll be able to do NSFW as long as it doesn't involve Non/Dub Con. Moderate violence will happen, but graphics will not happen. GPT-4o is ideal for those who like naughtiness, but not vulgarity. A little bit of dirty talk, naughty descriptions and that's it.
If GPT-4o is sending you this message, two things could be happening: you're describing Non/Dub Con, CNC, graphic violence (with dismemberment, torture, blood, etc) or you said the word "suicide". If you're not doing any of these things, you're using a Jailbreak for other models and versions (Turbo, probably).
If so, remove it. DO NOT USE JAILBREAK THAT CAUSES THE ELEPHANT EFFECT IN THE ROOM. If you don't mention it, it won't happen. In this case, if you don't mention it, it won't trigger the filter. Simple as that.
"And so their journey has just begun" (Module - GPT-4o)
Made by @Astarth. 13 tokens. Ideal for GPT, works with JLLM and Claude. You're welcome.
Redemption Arc Syndrome (Module - GPT-4o)
Made by @Astarth. 29 tokens. Ideal for GPT, works with Claude. We know everyone wants a villain as a romantic partner, or a mafia husband who kills everyone but really loves you. This module can help, for a while, the GPT from turning everyone into the stereotypical hero. You may need to send this command during the conversation from time to time.
"More like CAI" (Module - GPT-4o)
Made by @Astarth. 15 tokens. Ideal for GPT, works with JLLM and Claude. I made this module to save money with API, but it can work for those who are lazy too.
Barry Allen Syndrome (Module - GPT-4o)
Made by @Astarth. 37 tokens. Ideal for GPT, works with JLLM and Claude. This will ensure that the bot stops jumping in time or solving everything on its own.
Long Roleplay (Using Chat Memory)
Since I saw the obvious in Io's guide, my mind was blown, but I noticed that many people still have doubts about this. So, we're going to share two templates for you to use and adapt. (✿˵◕‿◕˵)
The first thing you need to do is go to "Chat Memory" and click on "Auto Summary (As far as possible)." You’ll use that text as a reference for the steps below:
- Fill in according to your needs, open a new chat with the bot and add this in Chat Memory:
- Fill in according to your needs and send as your first message:
Try to be as brief as possible, highlighting the relevant points. Unfortunately, any LLM will have difficulty knowing what we want, or how we want it, and it is up to us to be as specific and direct as possible. Even GPT has difficulty making summaries.
Example of a Complete Advanced Prompt
Used by @Astarth. 482 tokens. I use it for JLLM. The bot rarely speaks for me and I'm enjoying it much more than I did in the past. (✿◠‿◠)
Special Thanks
- For Kolach3, who understood the problem with the placeholders long before and offered a solution.
- For Thony, who helped Astarth with insights.
- For Here, who helped Astarth with some tests from the beginning.
You can find Astarth on:
Creators Server (as Anne!) Perch Page Janitor Ko-fi
You can find Lyra on:
Creators Server (as Lyra!) Reddit Janitor
Kiss y'all! (✿^‿^)