Drappier's Guide to NSFW Bots (Faraday/MythoMax Edition)

NOTE: THIS IS FOR MYTHOMAX 13B Q4_K_M. THIS MAY/MAY NOT WORK WELL WITH OTHER MODELS.

Simple Stuff

First, download Faraday.
Then download MythoMax 13B (This will probably work on lower-end and higher-end versions of the model, but I use Q4_K_M for my system.
After it's done downloading, go to your settings and set up a character name. You can also optionally add a default persona.

Click on "Create Character" and swap it to advanced.

DO NOT FUCK WITH MODEL INSTRUCTIONS.

Display name is the name of the AI when you go to click on it from the Home section.
Real name is used during ERP. You can keep it short to save tokens.

Character Persona is where you put the juice of the bot. Its how the bot will know what is it, who it is, how it looks, and how it acts. Usually, people use a style called W++ for this. To keep it as simple as possible, Ill just give you a W++ script I made for a bot as a reference.

1
2
3
4
5
6
7
8
9
[character()
{
species(draconcopode + monster girl)
mind(horny + bold + selfish + ferocious + animalistic + loving)
personality(horny + bold + selfish + ferocious + animalistic + loving)
body(humanoid + female + brown and yellow scaly skin + long snake tongue + yellow eyes + long scaly tail + huge hips + big ass + big breasts + tight vagina)
likes(sex + cuddle fucking + being pet + hugging with her her tail)
description(in a constant state of heat + rapes {user} on sight + will not stop sex until {user} has passed out + will cuddle {user})
}]

Here's a short breakdown of what I put above:

  • Character would be the name of the bot again, although you can keep it blank as the bot will remember its name.
  • Species matters if you're making a furry or monster girl
  • Mind is how the bot thinks and how the bot acts. Personality works the same way.
  • Body is how the character looks.
  • Likes is what the bot enjoys.
  • Description is a brief description of the bot. You can use it as a vague guide on how to act. It works if you split it up like above or if you do sentences.
    You can add more than that as well. Just look at https://github.com/KoboldAI/KoboldAI-Client/wiki/Pro-Tips#w for examples of what to put.

    Note: I believe "Dislikes" and "Hates" are a waste of tokens because it wont do what you tell it to do.

"Your Persona" I use as a brief description of my character, which is:

{user} is a goat and is also {character}'s play toy

Nothing too complex or you'll waste tokens.

Note: This will also probably be added automatically if you fill out the default persona.

Scenario is the same deal, just with a brief description of the greeting message. For this bot, I used:

{character} is a brown and yellow snake that found its way into {user}'s house.

Greeting message is the first message the bot sends to you. You typically wanna put a detailed scenario in this. It'll partially structure the sentences it outputs as well. In the case of the bot I was making, it should look something like this:

*You're fast asleep in your bed, dreaming something completely incoherent. In your dream, you feel a weird, coiling grip around your entire body. It starts to get tighter, causing you to wake up in panic. The moment you open your eyes, you see a yellow and brown snake gaze at you. She puts her right arm under her breast as she continues to coil you tighter. She gets closer to your face.*

"Hello, {user}... You're gonna be mine for the rest of your life..."

Example Messages is how the bot will structure their sentences. Usually, you want to put detailed RP here in a format like this:

<START>
#{user}:
#{character}:

User is something the person using it would say, and character would be how the bot generally responds. For example:

#{user}: G- Get off of me!
#{character}: *As you struggle, she takes off your clothes, adjusting her tail around you so she can remove them while keeping her grip on you. She takes her arms and puts them on your shoulders before kissing you deeply, playing with your tongue*

Generally, you'll want to make three or four generalized example messages unless you're making a linear scenario bot. MythoMax doesn't like to interpret example dialogue. Instead, it uses the example dialogues as a way to drive the scenario. If its generalized, though, its more likely to interpret it rather than completely using it.

Make sure you enable NSFW, your temperature is 0.8-1.0, and the Top K is 100. Don't touch ANYTHING else and just save the bot.

General tips for bot creation:

If you have no creative juices/cant think of a name/cant think of a scenario, you can try what I do: Get horny as fuck. Usually when I get horny, it'll make me think of scenarios and other shit I usually could never come up with.

When you make an example message, make sure to italicize the RP parts of it and put quotes around what the bot is supposed to say. Otherwise, it'll be more confusing to tell which part is RP and which part is the bot speaking.

Along with proper italicization and quotations, make sure to use stronger words. Don't put "I nut inside her pussy and moan loud" in your example messages. The detail is too little. Instead, you can do, "I do one final thrust before my cock erupts, pouring semen into her vagina violently. I moan loudly as this happens.", which will add more variety to the text upon generation.

Advanced Stuff/Detailed Explanations of My Bot(s)

Starting off this section, I think I need to address the elephant in the room:

Why is the reference I gave for the persona so low in tokens? Wouldn't you want to put more tokens into that than the example messages?

Well, you're partially right. In general, it'll depend on what you want from a bot. If you want it to have a very specific personality, then yeah, you'll probably want to put more into the personality than the example messages, but my method of making bots with a specific personality is somewhat weird. I make the example messages have dense RP with small lines of text so the bot speaks after. In the RP, I put relatively detailed words in it. For some reason, it'll shape the personality of the bot as well at .8 temperature. This also applies to the greeting, which is why I detail the scenario heavily and add relatively detailed RP related to the bot. You can see this in my non-con bots such as Wynter the Wyvern or Krampus. Also, adding less personalities allows people to somewhat control the bot's personality through roleplay. This is very prevalent with my Wendigo bot. Ask me for the PNG's on discord.)

So, now that I addressed that, what about the scenario? Why not put the greeting message in the scenario?

Although you can 100% do that, the main reason why I don't do that is because the reader wont know what's happening unless the greeting is also the scenario. If you detail the greeting message as hard as I do, you'll understand why I don't put it in the scenario. Most of the time, I don't even put a scenario cause it's somewhat pointless since the bot understands the scenario from the greeting message itself. Sometimes, though, I will put a very brief scenario in the bot's PoV to guide it. I usually do this for more complicated scenarios or when the bot doesn't know what the scenario is at all.

What makes your W++ style better than most?

Firstly, it optimizes tokens. Context size for MythoMax 13B is 2048, which is split in half: 1024 max for the bot, 1024 max for memory. 1024 tokens for memory, with my experience, will allow the bot to remember 4 outputs. Any bot that has less than 800 tokens will remember SO MUCH MORE. Secondly, the W++ format is a lot easier to understand than boostyle + Ali:Chat, which doesn't really work with Faraday's default settings since it usually requires using softprompts. I dont mind softprompts, but they shouldn't be required or "recommended" to run a bot. Making a bot run well with default settings will destroy any bot that requires "x" and "x" unless their responses are much better in comparison. Thirdly, not everyone is gonna modify Faraday to do very specific shit for them. Most users download it, click on a character, and use it. If you can reduce the amount of work on both ends and get a good output, then it doesn't matter. Last, but not least, my style allows for one-liner responses to the AI while the AI responds with a/multiple paragraph(s). This method is VERY powerful if done correctly.

"x" method is better.

I do not care. I've seen the outputs of each method and most perform the same EXCEPT for this one. AliChat takes too many tokens which means much less memory and boostyle gives very short responses. This method of W++ is simple to grasp, simple to make, and simple to work with.

Wanna critique something or have a general complaint about the guide? Contact me!
@angudaz on Discord

Edit
Pub: 02 Nov 2023 18:38 UTC
Views: 777