Skip to content

Formatting

When SillyTavern sends messages to IntenseRP, it doesn't just forward them straight to the active provider. Instead, it squishes everything into a single, neatly formatted prompt that the model can understand better. This page explains how that magic works.

Chat route only

This formatting pipeline applies to chat-style requests such as /v1/chat/completions.

If you use the legacy /v1/completions text-completions route, IntenseRP skips templates, injections, and name scanning there and forwards the raw prompt text instead.

The Big Picture

Here's what happens when you send a message:

flowchart LR
    A[SillyTavern Messages] --> B[Name Detection]
    B --> C[Apply Template]
    C --> D[Join with Divider]
    D --> E[Add Injection]
    E --> F[Send to Provider]
  1. Name Detection - Figure out who's talking (your character, the AI character, etc.)
  2. Apply Template - Format each message using your chosen template
  3. Join with Divider - Combine all messages with a separator (usually a newline)
  4. Add Injection - Optionally prepend or append extra instructions
  5. Send to Provider - Off it goes!

Templates

Templates control how each message looks. They use placeholders that get replaced with actual values.

Placeholders

Placeholder What It Becomes
{{name}} The character's name (e.g., "Sophie", "Jan")
{{role}} The role type: "User", "Character", or "System"
{{content}} The actual message text

Example

Let's say you have a message from your character named "Sophie" that says "Hello there!"

With the template {{name}}: {{content}}, it becomes:

Sophie: Hello there!

With <{{name}}>{{content}}</{{name}}>, it becomes:

<Sophie>Hello there!</Sophie>

Presets

Don't want to write your own template? That's fine! I wouldn't either. IntenseRP comes with several built-in presets you can choose from.

Preset dropdown in Settings

Classic

The default, simple format. Good for most use cases. Originally used by IntenseRP API.

Template: {{name}}: {{content}}

Example output: 

Sophie: Hello there!
Jan: Hey, what's up?

Template: {{role}}: {{content}}

Example output: 

Character: Hello there!
User: Hey, what's up?

XML-Like

Wraps messages in XML-style tags. Some models respond well to this structured format.

Template: <{{name}}>{{content}}</{{name}}>

Example output: 

<Sophie>Hello there!</Sophie>
<Jan>Hey, what's up?</Jan>

Template: <{{role}}>{{content}}</{{role}}>

Example output: 

<Character>Hello there!</Character>
<User>Hey, what's up?</User>

Multiline XML-Like

Almost exactly like XML-Like, but with the content placed between opening and closing tags on separate lines. This also matches the old v1 Wrapped/XML-like preset style.

Template: 

<{{name}}>
{{content}}
</{{name}}>

Example output: 

<Sophie>
Hello there!
</Sophie>
<Jan>
Hey, what's up?
</Jan>

Template: 

<{{role}}>
{{content}}
</{{role}}>

Example output: 

<Character>
Hello there!
</Character>
<User>
Hey, what's up?
</User>

Divided

Uses markdown-style headers to separate messages. Visually clear and easy to read.

Template: ### {{name}}\n{{content}}

Example output: 

### Sophie
Hello there!
### Jan
Hey, what's up?

Template: ### {{role}}\n{{content}}

Example output: 

### Character
Hello there!
### User
Hey, what's up?

Custom

Want to go wild? Select Custom and write your own template. Mix and match placeholders however you like!

Custom template editing

Message Divider

The divider is what goes between each formatted message. By default, it's just a newline (\n), but you can change it to anything.

Divider Result
\n (default) Messages on separate lines
\n\n Double-spaced messages
---\n Horizontal rules between messages
(space) All messages on one line (chaos mode)

Escape Sequences

Type \n to insert a newline. It'll be converted to an actual line break when formatting. You can just press Enter in the input box too, if you prefer.

Name Detection

IntenseRP tries to figure out who's who in your conversation. It looks for names in a few different places, depending on what your client sends.

Name Behavior settings

How It Works

IntenseRP checks these sources in order:

  1. Message Objects - The name field in each message (or legacy irp-next for RossAscends's STMP patcher compat)
  2. IR2 Blocks - Special tags in system messages
  3. Classic IntenseRP - Legacy DATA1/DATA2 format

If none of these work, it falls back to generic names: "User" and "Character".

Message Objects

The OpenAI API format lets you include a name field with each message:

{
  "role": "user",
  "name": "Sophie",
  "content": "Hello!"
}

When enabled, IntenseRP reads this name and uses it for {{name}} in your template.

Settings → Formatting → Name Behavior → Message Objects

STMP patcher compatibility (RossAscends's STMP)

If your client (like RossAscends's STMP) sends irp-next instead of name, IntenseRP treats it exactly like name (when Message Objects is enabled).

SillyTavern Setup Required

For SillyTavern to actually send these names, you need to configure it:

SillyTavern → User Settings → Character Names Behavior → Message

Without this, SillyTavern won't include the name field in messages, and IntenseRP won't have anything to detect.

Recommended

Using Message Objects is the most reliable way to pass character names. If your client supports it, always enable this option first.

IR2 Blocks

IR2 (short for "IntenseRP 2") is a simple, machine-readable way to embed character names in system messages. It's not a standard - just something I made up to reliably pass names through:

[[IR2u]]{{user}}[[/IR2u]]-[[IR2a]]{{char}}[[/IR2a]]

This tells IntenseRP that the user is "Sophie" and the AI character is "Jan".

Settings → Formatting → Name Behavior → IR2 Blocks

Classic IntenseRP

The original IntenseRP format from way back:

DATA1: "{{char}}" DATA2: "{{user}}"

Where DATA1 is the character name and DATA2 is the user name. (Yes, it's backwards. Don't ask.)

Settings → Formatting → Name Behavior → Classic IntenseRP

Priority

Methods are checked in order. If Message Objects finds a name, it won't bother checking IR2 or Classic. You can toggle each method on or off depending on what your client sends.

Injection

Sometimes you want to add a little extra instruction to every prompt, like a reminder to stay in character or follow certain rules. That's what injection does.

Injection settings

How It Works

Injection adds your custom text either before or after all other formatted messages.

Position Where It Goes
Before At the very beginning of the prompt
After At the very end of the prompt

Injection placeholders

Injection content supports these placeholders:

Placeholder Replaced with
{{user}} Detected user name
{{char}} Detected character name

The names come from the same Name Detection system (Message Objects, IR2 blocks, or Classic IntenseRP tags).

Example

If your injection content is:

[Remember: Stay in character at all times]

And your position is Before, the final prompt looks like:

[Remember: Stay in character at all times]
Sophie: Hello there!
Jan: Hey, what's up?

Setting It Up

  1. Choose your Position (Before or After)
  2. Enter your Content
  3. Save!

Use Cases

  • Add a "jailbreak" or instruction at the start
  • Append reminders about formatting or behavior
  • Insert context that should always be present

Disabling Formatting

If you want IntenseRP to pass messages through without any formatting, just turn it off:

Settings → Formatting → Apply Formatting → Off

When disabled, messages are formatted as simple role: content pairs:

user: Hello there!
assistant: Hey, what's up?

Quick FAQ

My character names aren't showing up?

Make sure at least one name detection method is enabled. If you're using SillyTavern, check that Character Names Behavior is set to Message in User Settings. Otherwise SillyTavern won't send the names at all.

Can I use multiple lines in my template?

Yes! Use \n for newlines in your template. For example: {{name}}\n{{content}} puts the name on its own line.

What if I just want raw messages?

Turn off Apply Formatting and IntenseRP will pass messages through with minimal processing.

What's the difference between IR2 and Classic IntenseRP?

IR2 is the newer format with XML-like tags ([[IR2u]]...[[/IR2u]]). Classic uses DATA1/DATA2 strings. Both do the same thing - they pass character names - but IR2 is cleaner and less likely to be accidentally parsed as content. It's deliberately designed to be written in a way that absolutely no normal user message would ever contain it.

Back to Features

Features Overview