Kounaisha's Immersive SillyTavern Guide for Windows

Kounaisha's Discord server Kounaisha's Ko-Fi page

Take a look at my other guides too:

How to use proxies on JanitorAI using OpenRouter - Jailbreak list for LLMs



Hello! I am Kounaisha and in this guide i'll teach you how to install, configure and use Silly Tavern, as well as creating bots and chatting with them.

What is Silly Tavern?

SillyTavern is a local hosted interface that allows you to create and interact with bots using LLMs, as well as enjoy TTS voice models, generate images, among many other interesting functions.
Silly Tavern is well-known mainly for its complexity (yes, it is VERY complex). As a JanitorAI user, I am already used to working with somewhat complex configurations, but I still have difficulty understanding certain functions. I will do my best to pass on to you everything I have learned on my own.

Okay but, like, What are the features that make SillyTavern so good?

Some of the main functions are:

  • A wide variety of settings for text generation;
  • World creation capability, i.e. you create world prompts and insert them into your characters;
  • Group chats: multi-bot rooms for characters to talk to you and/or each other;
  • UI customization options: theme colors, background images, custom CSS, and more;
  • User personas: let the AI know a bit about you for greater immersion;
  • Built-in RAG support: add documents to your chats for the AI to reference;
  • Extensive chat commands subsystem and own scripting engine;
  • Many, Many more!

How to Install SillyTavern on Windows

SillyTavern is written primarily in JavaScript and HTML and is open source. If you want to learn more about it before installing it, here is the github page: https://github.com/SillyTavern/SillyTavern. There are also installation instructions there, but I'll post them here anyway.

Requirements

Installing Silly Tavern

1 - Open Windows Explorer (Windows hotkey + E)
2 - Browse to or Create a folder that is not controlled or monitored by Windows. (ex: C:\YourFolder)
3 - Open a Command Prompt inside that folder by right clicking anywhere in that folder while pressing shift and then clicking the "Open Terminal" Option
4 - Once the terminal pops up, type the following into it and press Enter:
git clone https://github.com/SillyTavern/SillyTavern -b release
5 - Once everything is cloned, double-click Start.bat to make NodeJS install its requirements. The server will then start, and SillyTavern will pop up in your browser.


Setting up Silly Tavern

Connecting to an API

Now comes the real nerdy part! As soon as you start Silly Tavern, you will see this web page open in your browser:

The first thing we're going to do is set up our API connection. To do this, you can click on the plug icon in the navigation bar, or click where it says "API Connections"

A new window will open with several fields to fill in. The first one is the profile. By default, SillyTavern uses the "Horde" profile, but if you want, you can create another profile by clicking on the second icon next to it.

In the profile creation window, leave everything as is, just change the name in the text field and press "save"

Now let's configure the API. Click on the field that says "API" and select "Chat Completion"

In the "Chat Completion Source" field, you will select "OpenRouter"

Now comes the part where you need to go to OpenRouter's website, create an account and generate an API key. If you don't know how to do this, you can use my other guide for that. Follow the first steps there until you create an API key, then come back here.

After creating your API key, click the "authorize" button below. You will be redirected to the openrouter website asking for permission to authorize, along with a credit limit textfield. Just click the "authorize" button and you will be redirected back to sillytavern, with your key already saved.

SillyTavern will NEVER show your API key. If you want to change your key, simply click "authorize" again and repeat the process.

With the API connected, we can now select which LLM we will use. Click where it says "OpenRouter Model", and a huge list of available OpenRouter models will appear. Note that most of them have a usage price, but if you type "free" in the text field, the list of models will be filtered, showing only the free models.

  • You are free to choose any model you want. Some of them have more memory than others, but are more unstable. In my guide on how to use proxies in JanitorAI, I recommended several models with 128k memory tokens, so take a look if you want (I will create a separate guide to recommend models later)

One last piece of advice: make sure to uncheck any options under "Model Providers"

Done! Your API settings are ready to go!


Advanced Formatting

This is where things start to get a little more complicated. In this window, we will configure how LLM will communicate. If you are a JanitorAI user, you will have a little more ease in this part, but it still has some very advanced functions that it is recommended not to mess with if you do not know what you are doing.

Basically, there are three settings: Context Template, Instruct Template and System Prompt. They all have the same purpose of teaching LLM how to talk, but they do it in different ways. The "Instruct template", for example, uses different instruct methods depending on the model you are using, "Context Template" is how LLM will interpret your chat context, and "System Prompt" is basically our jailbreak.

For roleplay purposes, "System Prompt" is the only one really necessary here, but I will also use "Context Template".

Context Template

"Context Template" is the way LLM will "read" your chat history. This format varies from model to model, but it doesn't change the final result of the responses much. It basically serves as a way to prevent LLM from getting confused during reviews.

You can turn on Context Template by clicking on the button with the ⚡ icon and selecting a preset from the field right below. There are templates for various types of models. If by chance you don't find a template for the LLM you are using, use the "simple-proxy-for-tavern" template (or just leave it turned off).

System Prompt

This is where you will put your jailbreak. Each model is broken in a different way and there are several jailbreaks out there on the internet. Since I'm a good girl, I made a small list of jailbreaks of the most used AIs for roleplaying. just copy one of them and paste it into sillytavern, and you're done.

List of Jailbreaks

User Settings

User Settings are mostly visual/utility options, Including font size and formatting, theme type, among others. Nothing too important. If you want to take a look or want to customize your interface, go ahead, otherwise you can leave things as they are here.

Backgrounds

Another visual setting. You can choose from the available backgrounds or add a new one. Any image will do.

Extensions

Here things get a little more interesting. This tab contains all the extensions already installed in sillytavern, plus the function to add even more extensions. By default, sillytavern comes with very useful extensions, such as:

  • Character expressions (upload an image for each emotion. The image will appear in the left corner of the interface)
  • Image Generation (using both local and online interfaces)
  • TTS
  • Chat translation
  • Summarize
  • Many others

I won't go into too much detail about the extensions, but they are certainly very useful for increasing immersion.

If you are interested in adding more extensions, copy the link of one of these extensions, click the "install extensions" button in sillytavern's top right corner and paste the link into the text field.


Persona Management

Finally! In this tab, you can create your personas! And the best thing is that there is no limit to the number of personas! (as far as I know)

Creating a new persona

We'll need a persona to talk to the bots, so let's create one, shall we? In the persona tab, click the "create" button

A window will appear asking for a name. Choose the name of your choice for your persona and then click "save". Your persona is now created and you can complete it with an image, a description and a lore (yes, your personas can have LORE!).


Creating a character

Click on the ID icon in the navigation bar. A tab will appear on the right side. This tab is where your characters will be. To create a new one, simply click on the icon shown below:

The entire tab will be replaced by a character creation tab. In this tab, you will enter the name, tags (tags will be created if they do not exist), description (NOT personality), and initial message.

Now, to create the personality, we need to go to another window, clicking on the book icon.

A large window will appear in the center, where you can write the character's personality, scenario, level of conversation in groups (only in groups) and the example message. In addition, there are more advanced options, such as prompt overrides and post-story instructions (not necessary), as well as metadatas, in case you want to share your character with others and want to take credits for it.

When finished, click the confirmation icon to create the character.

(Tip: Whenever you are on a character tab, you can return to the character list by clicking on the three lines icon in the corner of the character creation tab)

when you finish creating the character, go back to the character list tab and click on your new character. the chat will appear in the center with the initial message you set. Just start typing and have fun!


Lorebooks

In sillytavern, you have the ability to create LORES for both your persona and your characters, as well as being able to create an entire world with this function! If you are interested in LORES, follow me!

Depending on the amount of tokens used, it may be necessary to use an LLM with a relatively high amount of memory. I recommend using one that has at least 32k memory tokens.

Click on the button with the book icon in the navigation bar (you thought I had skipped it, didn't you?) A small window will appear, called "Worlds/LOREbooks". You will click on the little "New" button below.

Choose a name of your choice and press "save". Your LOREbook is now created and selected. Its name will appear right next to the button we pressed just now.
Next to the name, there are some buttons we'll use to configure the LOREbook. Now click on the button with the + symbol

A text field will appear asking for an entry title. On the left side of the text field there is an arrow, click on it and more options will appear, similar to the window below:

So, how the hell this works?

Whenever the AI detects a word in a conversation that matches a keyword from a lorebook, it will understand that there is a specific concept for that, it will read what is in the lorebook description and use this understanding in it's responses.

For example: if you create a character who lives in a city called "X", you can create a lorebook with the keyword "X" and a description explaining in detail what the city is like. the AI will understand and relate that to the character.

The same thing can be done with more specific things, like the name of a pet, a valuable object, a person, ANYTHING can be written in a lorebook and it will be linked to the character and the story.

If you intend to create a specific personality or story for one of your personas, the correct way is to do it in a lorebook as well. just use {{user}} as a keyword

Connecting a persona to a lorebook

To connect a persona to a lorebook, simply go to the personas tab (the smiley face icon in the navigation bar), select the desired persona and click on the planet icon where the description text field is.

A window will open where you can select which lorebook you want to have a connection with that persona. If you have created a lorebook, select it here and click ok. Your persona will automatically be connected to that lorebook.

Connecting a character to a lorebook

The process of connecting a character to a lorebook is the same as for the persona, except that you have to go to the character list and click on the character you want to connect. Then, click on the planet icon that will be there.


Creating Chatgroups

Creating a chat group is almost the same as creating a character. Go to the character list and click on the icon shown below:

A tab will appear where you can create a name for the group, configure the order in which each character will respond, add the characters you want, among other functions.

Choose the configuration that suits you best. When you're ready, click the confirmation icon and your party will be created. The party will appear in the character list.

When you are chatting with a chatgroup, you will have an option called "Auto mode" in the group settings. This option measures, in seconds, the time in which each of the characters in the group will chat automatically. Cool, right?

You can also create a specific scenario for this group by clicking on the paper icon.

That's it! Now you have a group with multiple characters who talk to each other!


Generation Settings

Clicking on the first icon in the navigation bar will open a settings menu in the left corner.

Context Size (Memory Tokens)

The first option we see is the context size. In this case, it is the amount, in tokens, of LLM memory that you are currently using. You can manually change it to whatever value you want, and you can even unlock the bar by clicking the "Unlocked Context Size" box above (tokens can be up to 2 million), but I do not recommend doing this. Sillytavern automatically detects the amount of tokens that a model has in memory and updates the amount automatically here.

Max Response Length

Here you determine how many tokens you want in each response, maximum. The minimum amount is not defined here, so if you want to configure it, you must create a prompt for it in the system prompt.

Multiple Swipes per Generation

Here you determine how many different responses LLM will generate for the same dialog of yours. This is useful if you want to choose which response is best. Default is 1 response per generation.

Temperature and other settings

Next, we have the Temperature, Frequency penalty, Presence penalty and some other options that appear depending on which provider you are using. I suggest not changing the options named "min" and "top", since they are not that important for the LLM to work (I didn't understand a damn thing how they work).

Temperature

The Temperature, I believe you already know, changes the LLM's creativity. The default temperature is 1 and below that the LLM will generate more mechanical, less immersive and less creative responses. Raising the temperature above the default will make LLM generate more immersive responses, showing emotions, and acting on its own, but it can also generate completely nonsensical texts.

Frequency Penalty

The Frequency Penalty tries to prevent LLM from repeating the same prompts several times. The default value of the penalty is 0. Raising this value makes LLM generate more different responses, but it can also generate nonsensical texts or keep changing scenarios constantly.

Presence Penalty

The Presence Penalty is intended to prevent LLM from repeating the same phrases within the same prompt. Increase the value if you find yourself seeing your character saying the same things over and over again. (Yes, we will never have that old problem of JLLM repeating "I-I-I-I-I-I-I-I..." again.)

These are the settings in this section that I recommend configuring to your liking.

Utility Prompts

Seed

The seed value determines a specific generation result. Use if you want to maintain the same result across generations. Not recommended for roleplay.

I recommend not changing any of the settings below these.

Guide Updates

  • teach more about extensions
  • more (I forgot something for sure)
Edit
Pub: 16 Feb 2025 20:53 UTC
Edit: 11 Apr 2025 04:05 UTC
Views: 1025