Click slider numbers to input manually.
MAD LAB MODE ON
Kobold Presets
NovelAI Presets
Chat Completion Presets
Text Completion presets
Response (tokens)
Context (tokens)
Max prompt cost: –
AI Module
Changes the style of the generated text.
Temperature
Repetition Penalty
Rep Pen Range
Repetition Penalty Slope
Repetition Penalty Frequency
Repetition Penalty Presence
Min P
TFS
Top P
Top A
Top K
Mirostat Tau
Mirostat LR
Typical P
Linear
Quad
Conf
Min Length
Phrase Repetition Penalty
Preamble
Use style tags to modify the writing style of the output.
Banned Tokens
Sequences you don't want to appear in the output. One per line. Text or [token ids].
Logit Bias
Add
Helps to ban or reinforce the usage of certain tokens.
Unrestricted maximum value for the context size slider. Enable only if you know what you're doing.
Context Size (tokens)
Max Response Length (tokens)
Multiple swipes per generation
Middle-out Transform
Max prompt cost: Unknown
Max prompt cost: Unknown
Max prompt cost: Unknown
Display the response bit by bit as it is generated.
When this is off, responses will be displayed all at once when they are complete.
Temperature
Frequency Penalty
Presence Penalty
Top K
Top P
Repetition Penalty
Min P
Top A
Quick Prompts Edit
Main
Auxiliary
Post-History Instructions
Utility Prompts
Impersonation prompt
Prompt that is used for Impersonation function
World Info format template
Wraps activated World Info entries before inserting into the prompt. Use{0} to mark a place where the content is inserted.
Scenario format template
Use {{scenario}} to mark a place where the content is inserted.
Personality format template
Use {{personality}} to mark a place where the content is inserted.
Group Nudge prompt template
Sent at the end of the group chat history to force reply from a specific character.
New Chat
Set at the beginning of the chat history to indicate that a new chat is about to start.
New Group Chat
Set at the beginning of the chat history to indicate that a new group chat is about to start.
New Example Chat
Set at the beginning of Dialogue examples to indicate that a new example chat is about to start.
Continue nudge
Set at the end of the chat history when the continue button is pressed.
Replace empty message
Send this text instead of nothing when the text box is empty.
Seed
Set to get deterministic results. Use -1 for random seed.
Temperature
Top K
Top P
Typical P
Min P
Top A
TFS
Repetition Penalty
Rep Pen Range
Repetition Penalty Slope

Mirostat

Mode
Tau
Eta
Seed

GBNF Grammar

Samplers Order
Samplers will be applied in a top-down order. Use with caution.
Top K 0
Top A 1
Top P & Min P 2
Tail Free Sampling 3
Typical P 4
Temperature 5
Repetition Penalty 6
Samplers Order
Samplers will be applied in a top-down order. Use with caution.
Temperature 0
Top K Sampling 1
Nucleus Sampling 2
Tail Free Sampling 3
Top A Sampling 4
Typical P 5
Mirostat 8
Unified Sampling 9
Min P 10
Multiple swipes per generation
Temperature
Top K
Top P
Typical P
Min P
Top A
TFS
Epsilon Cutoff
Top nsigma
Min Keep
Eta Cutoff
Repetition Penalty
Rep Pen Range
Rep Pen Slope
Rep Pen Decay
Encoder Penalty
Frequency Penalty
Presence Penalty
No Repeat Ngram Size
Skew
Min Length
Maximum tokens/second

Smoothing Factor
Smoothing Curve

Threshold
Probability

Multiplier
Base
Allowed Length
Penalty Range
Sequence Breakers

Dynamic Temperature

Minimum Temp
Maximum Temp
Exponent

Mode
Tau
Eta

Contrastive Search

Banned Tokens/Strings
Global list
Preset-specific list

Logit Bias

Add
Helps to ban or reinforce the usage of certain tokens.

CFG

Scale
Negative Prompt

Grammar String

JSON Schema

Samplers Order
kcpp only. Samplers will be applied in a top-down order. Use with caution.
Top K 0
Top A 1
Top P & Min P 2
Tail Free Sampling 3
Typical P 4
Temperature 5
Repetition Penalty 6

Sampler Order

llama.cpp only. Determines the order of samplers. If Mirostat mode is not 0, sampler order is ignored.
Temperature
Top K
Top P
Typical P
Min P
Exclude Top Choices
DRY
Rep/Freq/Pres Penalties
Top N-Sigma

Sampler Priority

Ooba only. Determines the order of samplers.
Repetition Penalty
Presence Penalty
Frequency Penalty
DRY
Temperature
Dynamic Temperature
Quadratic / Smooth Sampling
Top Nsigma
Top K
Top P
Typical P
Epsilon Cutoff
Eta Cutoff
Tail Free Sampling
Top A
Min P
Mirostat
XTC
Encoder Repetition Penalty
No Repeat Ngram

Sampler Order

Aphrodite only. Determines the order of samplers.
DRY
Penalties
No Repeat Ngram
Dynatemp & Temperature
Top Nsigma
Top P & Top K
Top A
Min P
Tail-Free Sampling
Eta Cutoff
Epsilon Cutoff
Typical P
Cubic and Quadratic Sampling
XTC
Character Names Behavior ()
Continue Postfix ()
Continue sends the last message as assistant role instead of system message with instruction.
Combines consecutive system messages into one (excluding example dialogues). May improve coherence for some models.
Send the system prompt for supported models. If disabled, the user message is added to the beginning of the prompt.
Use search capabilities provided by the backend. Not free, adds a $0.02 fee to each prompt.
Allows using function tools. Can be utilized by various extensions to provide additional functionality. Not supported when Prompt Post-Processing with "no tools" is used!
Sends attached media in prompts if supported by the model. Videos must be less than 20 MB and under 1 minute long. Audio must be less than 20 MB.
Allows the model to return image attachments. Incompatible with the following features: function calling, web search, system prompt.
Allows the model to return its thinking process. This setting affects visibility only.
OpenAI-style options: low, medium, high. Minimum and maximum are aliased to low and high. Auto does not send an effort level.
Allocates a portion of the response length for thinking (min: 1024 tokens, low: 10%, medium: 25%, high: 50%, max: 95%), but minimum 1024 tokens. Auto does not request thinking.
Sets a dynamic reasoning depth level for thinking (Flash 3/Pro 3). High and low are supported by both, minimal and medium are Flash 3 only. Auto lets the model decide.
Allocates a portion of the response length for thinking (Flash 2.5/Pro 2.5) (min: 0/128 tokens, low: 10%, medium: 25%, high: 50%, max: 24576/32768 tokens). Auto lets the model decide.
Constrains the verbosity of the model's response.
Assistant Prefill
Assistant Impersonation Prefill
Prefills won't work when function calling is enabled and any tools are registered.
Logit Bias
Helps to ban or reinforce the usage of certain tokens. Confirm token parsing with Tokenizer.
View / Edit bias preset
Add bias entry
  Most tokens have a leading space.

API

Context: --, Response: --

API key

Get it here: Register (View my Kudos)
Enter 0000000000 to use anonymous mode.
For privacy reasons, your API key will be hidden after you click 'Connect'.

Models

Not connected...

API url

Example: http://127.0.0.1:5000/api
KoboldCpp works better when you select the Text Completion API and then KoboldCpp as a type!
Cancel
Not connected...
View hidden API keys

Advanced Formatting

Grayed-out options have no effect when Chat Completion API is used.

Context Template

Context Formatting

Instruct Template

Include Names

Instruct Sequences

Story String Sequences
User Message Sequences
User Message Prefix
User Message Suffix
Assistant Message Sequences
Assistant Message Prefix
Assistant Message Suffix
System Message Sequences
System Message Prefix
System Message Suffix
Misc. Sequences
First Assistant Prefix
Last Assistant Prefix
First User Prefix
Last User Prefix
System Instruction Prefix
Stop Sequence
User Filler Message

System Prompt

Custom Stopping Strings

JSON serialized array of strings

Tokenizer

Token Padding

Reasoning

Max
Reasoning Formatting
Prefix
Suffix
Separator

Miscellaneous

Non-markdown strings
Start Reply With

Worlds/Lorebooks

Active World(s) for all chats
Global World Info/Lorebook activation settings
Click to expand
Scan Depth
Context %
Budget Cap
Min Activations
Max Depth
Max Recursion Steps
Insertion Strategy
or

User Settings

Language:

UI Theme

Avatars:
Chat Style:
Media Style:
Notifications:
Theme Colors
Main Text
Italics Text
Underlined Text
Quote Text
Text Shadow
Chat Background
UI Background
UI Border
User Message
AI Message
Chat Width
Font Scale
Blur Strength
Shadow Width

Character Handling

Miscellaneous

Slow Fast

Custom CSS

Chat/Message Handling

# Msg. to Load
(0 = All)
Streaming FPS
Enter to Send:
Auto-swipe
Minimum generated message length Blacklisted words
Blacklisted word count to swipe
Auto-Continue
AutoComplete Settings

STscript Settings

Backgrounds

Chat backgrounds generated with the  Image Generation extension will appear here.

Extensions

(DEPRECATED) Extras API:

Not connected...

Persona Management

Usage Stats
+

Current Persona

[Persona Name]

Persona Description

Position

Tokens: 0

Connections

Global Settings

PNG
JSON
User Avatar
Chat
Character

text

- Advanced Definitions

Prompt Overrides (For Chat Completion and Instruct Mode)

Insert {{original}} into either box to include the respective default prompt from system settings.

Main Prompt

Tokens: counting...

Post-History Instructions

Tokens: counting...

Creator's Metadata (Not sent with the AI Prompt)

Everything here is optional

Created by

Character Version

Creator's Notes

Tags to Embed

Personality summary

Tokens: counting...

Scenario

Tokens: counting...

Character's Note

@ Depth

Role

Tokens: counting...

Talkativeness

How often the character speaks in group chats!
Shy Normal Chatty

Examples of dialogue

Important to set the character's writing style.
Tokens: counting...
Chat History

Select a World Info file for :

Primary Lorebook

A selected World Info will be bound to this character as its own Lorebook. When generating an AI reply, it will be combined with the entries from a global World Info selector. Exporting a character would also export the selected Lorebook file embedded in the JSON data.

Additional Lorebooks

Associate one or more auxillary Lorebooks with this character.
NOTE: These choices are optional and won't be preserved on character export!
☰
 entries
Comma separated (required) Primary Keywords
Logic
(ignored if empty) Optional Filter
Outlet Name
Scan Depth
Case-Sensitive
Whole Words
Group Scoring
Automation ID
Recursion Level
Inclusion Group
Group Weight
Sticky
Cooldown
Delay
Filter to Characters or Tags
Filter to Generation Triggers
Additional Matching Sources
☰
+++
☰
☰
${characterName}
Thought for some time
img1
img1 img2
img1 img2 img3
img1 img2 img3 img4

Welcome to SillyTavern!

Language:
SillyTavern is aimed at advanced users.

Looking for AI characters?

Import from supported sources or view Sample characters

Your Persona

Before you get started, you must select a persona name.
This can be changed at any time via the icon.

Persona Name:

Avatar
in this group
Go back

Alternate Greetings

Add
These will be displayed as swipes on the first message when starting a new chat. Group members can select one of them to initiate the conversation.
Click the button to get started!
Alternate Greeting #
Delete
Audio
0:00 / 0:00
Author's Note
Unique to this chat.
Checkpoints inherit the Note from their parent, and can be changed individually after that.
Tokens: 0

User inputs until next insertion: (disabled)
Character Author's Note (Private) Won't be shared with the character card on export.
Will be automatically added as the author's note for this character. Will be used in groups, but can't be modified when a group chat is open.
Tokens: 0
Default Author's Note
Will be automatically added as the Author's Note for all new chats.
Tokens: 0
Chat CFG
Unique to this chat.
Global CFG
Will be used as the default CFG options for every chat unless overridden.
CFG Prompt Cascading
Combine positive/negative prompts from other boxes.
For example, ticking the chat, global, and character boxes combine all negative prompts into a comma-separated string.

Token Probabilities
Select a token to see alternatives considered by the AI.
File Name File Size