Tool AI Dungeon Masters’ Corner

[whoremonger333]

Hi everyone,

This thread is dedicated to discussing methods, tricks, and deeper approaches for generating text with language models — with a focus on lewd, erotic, and NSFW storytelling.

The main goal here is to explore how to design and refine prompts, character structures, and narrative styles that help language models become better interactive storytellers — whether for solo story generation or for dynamic roleplay-like scenarios.

What we’ll cover:
• Prompt engineering for lewd/NSFW themes
• Creating consistent and believable characters with personality and memory
• Balancing tone, pacing, and consent in adult storytelling
• Tools and models for enhanced generation (e.g., local LLMs, soft prompts and parameters)
• Techniques to avoid repetition and generic output
• Using model feedback loops to refine style and themes
• Building multi-character scenes with believable dialogue
• Exploring taboo, kink, and psychological depth
• Abliterated LLMs for local use
• Cloud LLM chats interfaces and playgrounds that are not restricted

Disclaimer:

Creative Freedom with Responsibility: We encourage pushing the boundaries of AI storytelling and customization, but all content must remain within legal and ethical limits. Think of it as a sandbox with guardrails—explore, innovate, and experiment, but stay mindful of what’s right and respectful within the scope and legal system of your place of residence.

It’s about crafting adult narratives that are emotionally engaging, character-driven, and immersive. Whether you’re writing interactive fiction, developing NPCs for adult games, or experimenting with narrative AI in private projects, this is a space to share insights, questions, tools, and examples.

You're welcome to share your expertise, prompts et cetera.

 

[whoremonger333]

Ok. So first of all. The tooling for local LLM deployment.
I myself find LM studio very promising and fast evolving. Faster engine than ollama, native support for Apple's own MLX format and enough tools to fine tune the models.

In regards to models. I find Qwen3-30b-a3b to very good. Of course the abliterated (decensored) versions of it. Absolutely the best performance for 24GB RAM Apple Silicon.

Here's the list of models I fetched but didn't try all of them out yet.


Alibaba Qwen

qwen3-30b-a3b-abliterated-erotic-i1 IQ_3M (informaed quantization, wanna check if it performs as good or better that hybrid K_M)
qwen3-30b-a3b-abliterated-erotic Q3K_M
qwen3-30b-a3b-abliterated Q3K_M
qwen3-30b-a3b-abliterated-i1 i1-Q3_K_M
goekdeniz-guelmez_josiefied-qwen3-8b-abliterated-v1 Q8

Mistral
neuralbeagle14-7b Q_8

Llama
solar-1 0.7b-instruct-v1.0-uncensored — this one is great for building characters that are later fed as system prompts to other models (role play LLMs) and generating one off-shoot short stories.

In the cloud Qwen3-235B-A22B shines for me. I have seen it acting remarkable in multi-lingual settings, capable to split narrator from the character. Alas, no access to sampling settings hence all gets done at system prompt level.

 

[dvdsmith027]

I might sound noob (which I am probably), but can you post some video tutorial as well for folks like me ?

 

[whoremonger333]

You must be registered to see the links

apparently this is where you can grab and peek other ppls prompts.
it requires gmail or social media accounts to sign up that's I don't like it yet.

 

[whoremonger333]

I might sound noob (which I am probably), but can you post some video tutorial as well for folks like me ?

Sure, good start is

You must be registered to see the links

 

[whoremonger333]

And of course, totally forgot to tell where I get all the models —

You must be registered to see the links

You can grab official models trained by companies or the by credible soupmakers.
Like

You must be registered to see the links

,

You must be registered to see the links

 

[ayyyp]

Thanks for opening this thread, can't wait to experiment with this stuff. Kinda glad that I upgraded my PC recently, otherwise RAM would be an issue.
I hope it's okay with you if i just repost your example prompt here.

Spoiler: example prompt

You don't have permission to view the spoiler content. Log in or register now.

 

[whoremonger333]

Later on we will discuss such things as sampling parameters. They play a massive role in outputs that you get. While system prompts are very important, it's only the parameters that can truly diversify the narrative.

 

[Doner077]

Hey we wanted to keep this convo focused on LLMs right? Not image/video generation models

 

[whoremonger333]

No, I don't think we're limited to text only.

 

[randomgamer986]

Hey we wanted to keep this convo focused on LLMs right? Not image/video generation models

please do NOT spam with gen art. we can talk about image gen software tho.

i myself have an amd automatic1111 with zeluda running

 

[Doner077]

please do NOT spam with gen art. we can talk about image gen software tho.

i myself have an amd automatic1111 with zeluda running

I didnt do anything man xd

 

[randomgamer986]

I didnt do anything man xd

sorry didn't mean to come off aggressive. its just really annoying when ppl flood with LQ ai gen art like its the coolest thing since sliced bread.

 

[cabritini]

Do you guys know how do i make the AI behave like AI Dungeon? a prompt or something, like, where can i give the AI informations about the story?

 

[whoremonger333]

Do you guys know how do i make the AI behave like AI Dungeon? a prompt or something, like, where can i give the AI informations about the story?

System prompt tells the instruction tuned model to follow your rules. Oftentimes you won't have the permissions to set it as there's mostly developer centric platforms that allow this.

So in this case you instruct the AI to do what you want it to do in your initial message.

 

[randomgamer986]

Do you guys know how do i make the AI behave like AI Dungeon? a prompt or something, like, where can i give the AI informations about the story?

This really depends on what you are using. The best way is to look at what others did on the platform you are using.

 

[ayyyp]

Ok guys what i did might be lazy might cost a few bucks but the free alternative is also more than enough i think.

Some of you might know

You must be registered to see the links

already but i didn't and only found it recently, basically it is character.ai for porn but completely free as well.

You can simply make an account look for what you like through a "tags" system and without any setup you can just use their own LLM for text generation, where you can even set the system prompt (I'll attach one that is very generally worded).

You can also use any other LLM you want through proxy, what makes it so easy is that you just do the setup once with it being extremely easy to try and switch out LLMs.
Easy Tutorial for proxy setup:

You must be registered to see the links

Alright have fun, happy gooning this will make you addicted i swear, there is so much variance so many characters that you can try out and all of it for free.

 

[ayyyp]

I can also recommend DeepSeek V3 0324, it has like no filter and can even straight up call you slurs if you're into that.

 

[ayyyp]

I am currently using this guides prompts, specifically Base + NSFW, ill add the formatting part now as well.

This is of course focused on DeepSeek so i dont know how effective it is with other LLMs, but i use the same prompt for the normal JanitorLLM and that works pretty good too.

You must be registered to see the links

 

[whoremonger333]

I can also recommend DeepSeek V3 0324, it has like no filter and can even straight up call you slurs if you're into that.

It feels like DeepSeek with chutes.ai is no longer available for free. The information about 200 calls per day seems untrue as well.