KatherineWAR Custom Prompt


Hey guys!My name is Katherine. I'm a bot author on Janitor AI and I also write prompts.I'm a huge fan of roleplay and have spent a long time trying to achieve the best possible quality of text responses from AI. My priorities have always been realism and logic, especially in the bot's reactions and dialogues.I have created several prompts that cover the most important aspects of roleplay. The prompts vary in size by tokens, so keep that in mind when choosing.I strive for quality and for actually feeling something from the roleplay — to get positive emotions and simply relax, unwind my mind with an interesting story with AI. And I wish the same for everyone.Since I was previously a roleplayer with real people, I understand how an interesting literary roleplay should be built. I value both the quality of the text and its volume and content. And in general I can be quite a stickler about it. :DYes, the prompts are very large. Please don't write to me about this fact. It was done intentionally, because a small amount of information in a prompt has shown itself to perform extremely poorly in practice. And I would rather create and use a large prompt than read yet another piece of nonsense from AI.The prompts have been tested on the following models:
Qwen
Kimi
GLM
DeepSeek
I continue testing, continue looking for writing approaches that will be even better perceived by AI, as much as possible in text format.


Settings and Recommendations

The additional settings in Janitor (and on other platforms where they are available) are really useful.Important to understand! It won't be perfect, it will be better yes, but without your guidance the model will drift one way or another. Unfortunately the world's brightest minds and developers haven't yet figured out how to prevent information from being overwritten in favor of new data. This is called the "context window problem." So you should keep this in mind in general when using any AI for any purpose.AI is a tool and it is not yet perfect.Below are the ranges and screenshots with my settings.

ParameterSettings
TemperatureGood range: 0.9 to 1.5
Top-KGood range: 20 to 60
Top-PGood range: 0.6 to 0.9

Also, don't forget about OOC (Out Of Character).
Using OOC you are free to set the plot, direction, clarifications, explanations and anything else you wish. These are instructions for the AI outside of the roleplay.
What OOC is useful for:Correcting the AI on the fly:
- The AI went off track — you write [OOC: return to the previous scene]
- The bot forgot a detail — [OOC: remember that my character has black hair]
- The tone shifted wrong — [OOC: write more seriously, less humor]
Controlling the narrative:
- [OOC: move the plot forward, skip to the next morning]
- [OOC: introduce a new character into the scene]
- [OOC: end this scene and start a new location]
Adjusting the writing style:
- [OOC: write longer responses with more detail]
- [OOC: add more internal thoughts of the characters]
- [OOC: less dialogue, more action and description]
Technical instructions:
- [OOC: do not break character in your next response]
- [OOC: focus on the NPC's reaction, not mine]
The key rule: OOC should be short and specific. The more precise your instruction, the better the AI will follow it. Think of it as a director giving a quick note to an actor between takes — brief, clear, and to the point.


FAQ

Q: Is it normal that the prompts are so large?
A: Yes. I designed them that way. Short versions of my prompts and other small third-party prompts perform significantly worse in practice. At least for me personally. So I chose the path of providing more information to the AI.
Q: Will the prompts work with other AI models besides DeepSeek, Qwen, Kimi, and GLM?
A: Yes. The prompts are quite universal. Test them yourself.
Q: I can't copy your prompt!
A: This won't work on mobile. The prompts are large and the mobile clipboard simply won't be enough. Copy from a PC.
Q: I can't save the prompt to my proxy profile. Error 413.
A: The prompt tokens adjust to the tokens of the selected context. If an error occurs, it means you are exceeding the limit and you need to clear your proxy profiles and free up tokens.
Q: Will your prompts work on other platforms besides Janitor AI?
A: Yes. If it concerns setting up a proxy profile.
Q: Does the prompt size affect response speed?
A: Response speed does not depend on the prompt. It depends on your provider and the degree of model load. The only exception is thinking models — reading the prompt can take a bit longer due to its size. But personally, I haven't found it to be a major issue.
Q: Can I delete/add something to your prompts?
A: Go ahead and try. I created them over a long period of time with many tests. If you modify anything, I cannot guarantee quality responses from the AI. Either way, you are free to experiment.
⚠️A personal warning from me.
Be careful with services like OpenRouter where a pay-as-you-go payment model is available.
The prompts are large, and this may affect your spending.


My Custom Prompt


You can contact me, leave a review or ask a question:
Discord
Janitor AI (comments)